The Surface Science of Titanium Dioxide

Surface Science Reports 48 (2003) 53±229

The surface science of titanium dioxide Ulrike Diebold* Department of Physics, Tulane University, New Orleans, LA 70118, USA Manuscript received in final form 7 October 2002

Abstract

Titanium dioxide is the most investigated single-crystalline system in the surface science of metal oxides, and the literature on rutile (1 1 0), (1 0 0), (0 0 1), and anatase surfaces is reviewed. This paper starts with a summary of the wide varietyof technical ®elds where TiO 2 is of importance. The bulk structure and bulk defects (as far as relevant to the surface properties) are brie¯yreviewed. Rules to predict stable oxide surfaces are exempli®ed on rutile (1 1 0). The surface structure of rutile (1 1 0) is discussed in some detail. Theoreticallypredicted and experimentallydetermined relaxations of surface geometries are compared, and defects (step edge orientations, point and line defects, impurities, surface manifestations of crystallographic shear planesÐCSPs) are discussed, as well as the image contrast in scanning tunneling microscopy(STM). The controversyabout the correct model for the (1 Â 2) reconstruction appears to be settled. Different surface preparation methods, such as reoxidation of reduced crystals, can cause a drastic effect on surface geometries and morphology, and recommendations for preparing different TiO2(1 1 0) surfaces are given. The structure of the TiO2(1 0 0)-(1 Â 1) surface is discussed and the proposed models for the (1 Â 3) reconstruction are criticallyreviewed. Veryrecent results on anatase (1 0 0) and (1 0 1) surfaces are included.

The electronic structure of stoichiometric TiO2 surfaces is now well understood. Surface defects can be detected with a varietyof surface spectroscopies. The vibrational structure is dominated bystrong Fuchs±Kliewer phonons, and high-resolution electron energyloss spectra often need to be deconvoluted in order to render useful information about adsorbed molecules. The growth of metals (Li, Na, K, Cs, Ca, Al, Ti, V, Nb, Cr, Mo, Mn, Fe, Co, Rh, Ir, Ni, Pd, Pt, Cu, Ag, Au) as well as some metal oxides on TiO2 is reviewed. The tendencyto `wet' the overlayer, the growth morphology,the epitaxial relationship, and the strength of the interfacial oxidation/reduction reaction all follow clear trends across the periodic table, with the reactivity of the overlayer metal towards oxygen being the most decisive factor. Alkali atoms form ordered superstructures at low coverages. Recent progress in understanding the surface structure of metals in the `strong-metal support interaction' (SMSI) state is summarized.

Literature is reviewed on the adsorption and reaction of a wide varietyof inorganic molecules (H 2,O2,H2O, CO, CO2,N2,NH3,NOx, sulfur- and halogen-containing molecules, rare gases) as well as organic molecules (carboxylic acids, alcohols, aldehydes and ketones, alkynes, pyridine and its derivates, silanes, methyl halides).

* Tel.: 1-504-862-8279; fax: 1-504-862-8702. E-mail address: [email protected] (U. Diebold).

The application of TiO2-based systems in photo-active devices is discussed, and the results on UHV-based photocatalytic studies are summarized.

Keywords: Titanium oxide; Scanning tunneling microscopy; Single-crystalline surfaces; Adhesion; Catalysis; Chemisorption; Epitaxy; Growth; Interface states; Photochemistry; Surface relaxation and reconstruction; Surface structure; Morphology; Roughness; Topography

Contents

1. Introduction ...... 57 1.1. Motivation ...... 57

1.2. Applications of TiO2...... 59 1.3. Outline of this review ...... 64

2. The structure of TiO2 surfaces ...... 65 2.1. Bulk structure ...... 66 2.1.1. Bulk defects ...... 68

2.2. The structure of the rutile TiO2(1 1 0) surface ...... 70 2.2.1. The (1Â1) surface ...... 70 2.2.1.1. Bulk truncation ...... 70 2.2.1.2. Relaxations ...... 72 2.2.1.3. Appearance in STM and AFM ...... 74 2.2.1.4. Surface defects ...... 78 2.2.1.4.1. Step edges ...... 78 2.2.1.4.2. Oxygen vacancies created by annealing ...... 81 2.2.1.4.3. Oxygen vacancies created by other means ...... 84 2.2.1.4.4. Line defects ...... 84 2.2.1.4.5. Impurities ...... 84 2.2.1.4.6. Crystallographic shear planes ...... 85 2.2.2. Reconstructions ...... 88 2.2.2.1. Reconstruction under reducing conditions: the structure(s) of the (1Â2) phase ...... 88 2.2.2.2. Restructuring under oxidizing conditions ...... 89 2.2.3. Recommendations for surface preparation ...... 92 2.3. The structure of the rutile (1 0 0) surface ...... 93

2.3.1. The TiO2(1 0 0)-(1 Â 1) surface ...... 93 2.3.2. Reconstructions ...... 95

2.3.2.1. The microfacet model of the rutile TiO2(1 0 0)-(1Â3) surface...... 95 2.3.2.2. Is the simple microfacet model valid? ...... 96 2.4. Rutile (0 0 1)...... 96 2.5. Vicinal and other rutile surfaces ...... 99 2.6. Anatase surfaces...... 99 2.6.1. Anatase (1 0 1) ...... 100 2.6.2. Anatase (0 0 1) ...... 102 2.6.3. Other anatase surfaces ...... 103 2.7. Conclusion...... 103 U. Diebold / Surface Science Reports 48 (2003) 53±229 55

3. Electronic and vibrational structure of TiO2 surfaces...... 105 3.1. Stoichiometric TiO2 surfaces ...... 105 3.2. Reduced TiO2 surfaces ...... 109 3.2.1. Defect states ...... 109 3.2.2. Band bending ...... 110 3.2.3. Identi®cation of the reduction state with spectroscopic techniques...... 110 3.3. Vibrational structure ...... 111

4. Growth of metal and metal oxide overlayers on TiO2 ...... 112 4.1. Overview and trends ...... 112 4.1.1. Interfacial reactions ...... 112 4.1.2. Growth morphology(thermodynamic equilibrium)...... 115 4.1.3. Growth kinetics, nucleation, and defects...... 121 4.1.4. Film structure and epitaxial relationships ...... 122

4.1.5. Thermal stabilityof metal overlayerson TiO 2-SMSI ...... 122 4.1.6. Chemisorption properties ...... 124

4.2. Metals and metal oxides on TiO2 ...... 124 4.2.1. Lithium ...... 124 4.2.2. Sodium ...... 124 4.2.3. Potassium ...... 125 4.2.4. Cesium...... 126 4.2.5. Calcium ...... 127 4.2.6. Aluminum ...... 127 4.2.7. Titanium...... 127 4.2.8. Hafnium...... 128 4.2.9. Vanadium ...... 128 4.2.10. Vanadia ...... 129 4.2.11. Niobium...... 130 4.2.12. Chromium ...... 132 4.2.13. Molybdenum...... 132 4.2.14. Molybdena ...... 133 4.2.15. Manganese ...... 133 4.2.16. Manganese oxide...... 133 4.2.17. Iron ...... 133 4.2.18. Ruthenium ...... 135 4.2.19. Ruthenium oxide ...... 135 4.2.20. Cobalt ...... 135 4.2.21. Rhodium ...... 136 4.2.22. Iridium...... 137 4.2.23. Nickel ...... 137 4.2.24. Palladium ...... 138 4.2.25. Platinum...... 139 4.2.26. Copper ...... 142 4.2.27. Silver ...... 143 4.2.28. Gold ...... 144 4.3. Conclusion...... 147

5. Surface chemistryof TiO 2 ...... 148 5.1. Inorganic molecules ...... 148 56 U. Diebold / Surface Science Reports 48 (2003) 53±229

5.1.1. Hydrogen ...... 148 5.1.2. Water ...... 148 5.1.3. Oxygen ...... 155 5.1.4. Carbon monoxide and carbon dioxide ...... 156 5.1.4.1. CO...... 156

5.1.4.2. CO2 ...... 159 5.1.5. Nitrogen-containing molecules (N2, NO, NO2,N2O, NH3)...... 159 5.1.5.1. N2 (Table 12) ...... 159 5.1.5.2. NO...... 161

5.1.5.3. N2O...... 161 5.1.5.4. NO2 ...... 161 5.1.5.5. NH3 ...... 163 5.1.6. Sulfur-containing molecules (SO2,H2S, Sn)...... 163 5.1.6.1. SO2 ...... 163 5.1.6.1.1. TiO2(110)...... 163 5.1.6.1.2. TiO2(100)...... 164 5.1.6.2. H2S...... 165 5.1.6.3. Elemental sulfur (Sn, n 2)...... 165 5.1.7. Halogen-containing molecules (Cl2, CrO2Cl2,HI)...... 167 5.1.7.1. Cl2 ...... 167 5.1.7.2. Other halogen-containing molecules...... 169 5.1.8. Rare gases (Ar, Xe) ...... 170 5.2. Adsorption and reaction of organic molecules ...... 170 5.2.1. Carboxylic acids (formic acid, acetic acid, propanoic acid, acrylic acid, benzoic acid, bi-isonicotinic acid, oxalic acid, glycine, maleic anhydride) ...... 179 5.2.1.1. Formic acid (HCOOH) ...... 179 5.2.1.2. Formate: adsorption geometryand structure ...... 180

5.2.1.2.1. TiO2(1 1 0)-(1Â1)...... 180 5.2.1.2.2. TiO2(1 1 0)-(1Â2)...... 181 5.2.1.2.3. Modi®ed TiO2(1 1 0) surfaces...... 181 5.2.1.2.4. Other TiO2 surfaces ...... 183 5.2.1.2.5. Anatase ...... 183 5.2.1.3. Reaction of formic acid ...... 183 5.2.1.4. Formic acidÐconclusion ...... 187

5.2.1.5. Acetic acid (CH3COOH) ...... 187 5.2.1.6. Propanoic acid (C2H5COOH) ...... 189 5.2.1.7. Acrylic acid (CH2=CHCOOH) ...... 189 5.2.1.8. Benzoic acid (C6H5COOH)...... 189 5.2.1.9. Bi-isonicotinic acid ...... 189 5.2.1.10. Oxalic acid (HOOC±COOH)...... 190

5.2.1.11. Glycine (NH2CH2COOH)...... 190 5.2.1.12. Maleic anhydride ...... 191 5.2.2. Alcohols (methanol, higher alcohols) ...... 191 5.2.2.1. Methanol ...... 191

5.2.2.1.1. Methanol on TiO2(110)...... 192 5.2.2.1.2. Methanol on TiO2(0 0 1) and TiO2(100)...... 192 5.2.2.2. Higher alcohols ...... 194 U. Diebold / Surface Science Reports 48 (2003) 53±229 57

5.2.3. Aldehydes (RCHO) and ketones (RCOCH3) (formaldehyde, acetaldehyde, benzaldehyde, acetone, acetophenone, p-benzoquinone, cyclohexanone, cyclohexenone) ...... 194 5.2.3.1. Formaldehyde ...... 195 5.2.3.2. Acetaldehyde...... 195 5.2.3.3. Benzaldehyde ...... 196 5.2.3.4. Acetone and acetophenone ...... 196 5.2.3.5. Cyclic ketones...... 196

5.2.4. Cyclo-trimerization of alkynes (RCBCH) on reduced TiO2 surfaces and related reactions ...... 196 5.2.5. STM of pyridine, its derivates, and other aromatic molecules (pyridine, 4-methylpyridine, benzene, m-xylene, phenol) ...... 198

5.2.6. Adsorption and reaction of silanes (RSiX3) (TEOS, diethyldiethoxysilane, vinyltriethoxysilane, aminopropyltriethoxysilane, (3,3,3-tri¯uoropropyl)-trimethoxysilane) ...... 199

5.3. Photocatalysis on TiO2 ...... 200 5.3.1. Heterogeneous photocatalysis ...... 201 5.3.2. Photovoltaic cells ...... 202

5.3.3. Photocatalysis on single-crystalline TiO2 ...... 204 5.3.3.1. Oxygen, water, CO, and CO2 ...... 204 5.3.3.2. Alcohols ...... 205

5.3.3.3. CHX3 (X Cl,Br,I)...... 205 6. Summaryand outlook ...... 206 6.1. What has been learned and what is missing? ...... 206

6.2. TiO2 in relation to other transition metal oxides...... 207 6.3. TiO2Ðmixed and doped ...... 209 6.4. Nanostructured TiO2 ...... 209 6.5. Going beyond single crystal and UHV studies ...... 211 6.6. Concluding remarks ...... 212 Acknowledgements ...... 212 References ...... 212

1. Introduction

1.1. Motivation

The surface science of metal oxides is a relatively young ®eld that enjoys a rapidly increasing interest. The general trend to take the `next step' in surface scienceÐto move on to more realistic and complex model systemsÐlets many researchers to develop an interest in oxide surfaces. This is motivated bythe desire to contribute to the numerous applications where oxide surfaces playa role; after all, most metals are oxidized immediatelywhen exposed to the ambient. The knowledge of well-characterized single-crystalline metal oxide surfaces is reviewed extensively byHenrich and Cox [1] in 1993. This excellent book (which has become a classic in the ®eld) starts by showing the number of publications per year on fundamental surface-science studies on all metal oxides. The number of papers culminates with around 100 articles in 1991, the last year reviewed. A 58 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 1. Number of publications on single-crystalline TiO2 surfaces/year. Courtesy of M.A. Henderson, Paci®c Northwest National Laboratory.

similar analysis (Fig. 1) of (experimental) papers on single-crystalline TiO2 surfaces shows that more than 70 articles were published on the TiO2(1 1 0) surface alone in the year 2000. What is the reason for the popularityof this system?One driving force for pursuing research on single-crystalline TiO2 surfaces is the wide range of its applications and the expectation that insight into surface properties on the fundamental level will help to improve materials and device performance in many®elds. Titanium dioxide is a preferred systemfor experimentalists because it is well-suited for manyexperimental techniques. Polished crystalswith a high surface qualitycan be purchased from various vendors. Theycan be reduced easily,which convenientlyprevents charging of this wide band gap semiconductor. One also should not underestimate the `self-promoting' effect of popularityÐnew phenomena are studied most easilyon well-characterized, often tested systems,and TiO 2, especiallythe most stable rutile (1 1 0) surface, falls certainlyinto this category.All these factors have contributed in making TiO2 the model system in the surface science of metal oxides. Despite this high interest, a comprehensive review of the surface science of TiO2 is lacking at this point. Several excellent reviews of different aspects of single-crystalline metal oxide surfaces were written in recent years [1±10], and TiO2 surfaces are considered in almost all of them. Still, the time maybe ripe to review the wealth of knowledge on TiO 2 itself, and an attempt is made in this paper. It is intended to give the interested reader an introduction into TiO2, and clarifysome confusing and con¯icting results, e.g. on the structure of TiO2 surfaces as observed with scanning tunneling microscopy(STM), the adsorption of test molecules such as water and formic acid, and the rich bodyof literature on metal growth on TiO2 surfaces. There is also a hope that the insights obtained on this model oxide can be transferred, at least in part, to other systems. The focus is on the more recent literature (>1990). While an attempt was made to include most of the single-crystalline work on TiO2 U. Diebold / Surface Science Reports 48 (2003) 53±229 59 surfaces, the sheer number of papers excludes comprehensiveness, and apologies are extended to any authors whose work was unfortunatelynot represented.

1.2. Applications of TiO2

Before dwelling on actual surface science results, a brief glimpse on the applications of TiO2 (which, after all are the deeper motivation for most of the performed work) is in order. Titanium dioxide is used in heterogeneous catalysis, as a photocatalyst, in solar cells for the production of hydrogen and electric energy, as gas sensor, as white pigment (e.g. in paints and cosmetic products), as a corrosion-protective coating, as an optical coating, in ceramics, and in electric devices such as varistors. It is important in earth sciences, plays a role in the biocompatibility of bone implants, is being discussed as a gate insulator for the new generation of MOSFETS and as a spacer material in magnetic spin-valve systems, and ®nds applications in nanostructured form in Li-based batteries and electrochromic devices. A better understanding and improvement of catalytic reactions is one main driving force for surface investigations on TiO2. Because most heterogeneous catalysts consist of small metal clusters on an oxide support, manygrowth studies of metals on TiO 2 were performed. These metal/TiO2 systems often serve as a model for other metal/oxide surfaces. Traditionally, TiO2 is a component in mixed vanadia/ titania catalysts used for selective oxidation reactions [11]. The surface science of vanadium and vanadia/TiO2 systems was addressed by several groups [12±15].TiO2 is not suitable as a structural support material, but small additions of titania can modifymetal-based catalystsin a profound way.The so-called strong-metal support interaction (SMSI) is, at least in part, due to encapsulation of the metal particles byan reduced TiO x overlayer (see review by Haller and Resasco [16]). Recently, this phenomenon was revisited using surface science techniques [17±20]. The discoverythat ®nely dispersed Au particles supported on TiO2 and other reducible metal oxides oxidize CO at low temperature [21] has spurred some excitement in the surface science community. Many experiments that mayclarifythe underlyingphenomena leading to this processes are still underway [22±24].

The photoelectric and photochemical properties of TiO2 are another focus of active research. The initial work byFujishima and Honda [25] on the photolysis of water on TiO2 electrodes without an external bias, and the thought that surface defect states mayplaya role in the decomposition of water into H2 and O2, has stimulated much of the earlywork on TiO 2 [26±28]. Unfortunately, TiO2 has a low quantum yield for the photochemical conversion of solar energy. The use of colloidal suspensions with the addition of dye molecules has been shown to improve ef®ciency of solar cells [29], and has moved

TiO2-based photoelectrochemical converters into the realm of economic competitiveness [30]. Byfar, the most activelypursued applied research on titania is its use for photo-assisted degradation of organic molecules. TiO2 is a semiconductor and the electron±hole pair that is created upon irradiation with sunlight mayseparate and the resulting charge carriers might migrate to the surface where theyreact with adsorbed water and oxygento produce radical species. These attack anyadsorbed organic molecule and can, ultimately, lead to complete decomposition into CO2 and H2O. The applications of this process range from puri®cation of wastewaters [31]; desinfection based on the bactericidal properties of TiO2 [32] (for example, in operating rooms in hospitals); use of self-cleaning coatings on car windshields [33], to protective coatings of marble (for preservation of ancient Greek statues against environmental damage [34]). It was even shown that subcutaneous injection of a TiO2 slurryin rats, and subsequent near-UV illumination, could slow or halt the development of tumor cells

[35±37]. Several review papers discuss the technical and scienti®c aspects of TiO2 photocatalysis 60 U. Diebold / Surface Science Reports 48 (2003) 53±229

[31,38±42]. An extensive review of the surface science aspects of TiO2 photocatalysis has been given byLinsebigler et al. [43], and some of these more recent results are discussed in Section 5.3.3. Semiconducting metal oxides maychange their conductivityupon gas adsorption. This change in the electrical signal is used for gas sensing [44].TiO2 is not used as extensivelyas SnO 2 and ZnO, but it has received some attention as an oxygen gas sensor, e.g. to control the air/fuel mixture in car engines

[45,46]. Two different temperature regimes are distinguished [47]. At high temperatures, TiO2 can be used as a thermodynamically controlled bulk defect sensor to determine oxygen over a large range of partial pressures. The intrinsic behavior of the defects responsible for the sensing mechanism can be controlled bydoping with tri- and pentavalent ions. At low temperatures, addition of Pt leads to the formation of a Schottky-diode and a high sensitivity against oxygen [47].

The sheer volume of TiO2 pigments produced world-wideÐcurrentlyca. 4 million tons per yearÐis stunning [48].TiO2 pigment is used in virtuallyeverykind of paint because of its high refractive index. (See Table 1 for a summaryof bulk properties of TiO 2. A more detailed resource on rutile was given in [49].) The surface properties playa role even in these wide-spread applications, e.g. the photocatalytic degradation of binder in paints is a major problem for the paint industry. TiO2 is non-toxic and safe, and can be dispersed easily [48]. In pure form it is also used as a food additive [50], in pharmaceuticals, and in cosmetic products [51]. Titanium dioxide is used extensivelyin thin-®lm optical-interference coatings [52]. Such coatings are based on the interference effects between light re¯ected from both the upper and lower interface of a thin ®lm. (The same effect gives rise to the different colors of an oil ®lm on water.) The relative ratios between transmission and re¯ection of light are governed bythe index of refraction of the thin ®lm and the surrounding media. Bydepositing a stack of layerswith the appropriate optical index, the refraction/transmission properties of a stack of thin layers on a glass substrate can be designed to meet a great number of applications. Examples for such devices include antire¯ective coatings, dielectric mirrors for lasers, metal mirrors with enhanced re¯ection, and ®lters [52]. For most ®lms a combination of materials with indices as high and as low as possible is an advantage. Titanium dioxide has the highest index of all oxides (see Table 1), making it ideallysuited for this application. One of the `hot' issues currentlydebated in materials science is the search for the best dielectric gate material for replacing SiO2 MOSFET devices [53]. It appears that the limit for miniaturization, when electric tunneling through ever thinner SiO2 ®lms becomes signi®cant, will be reached in the verynear future. Ultrathin metal oxide ®lms might be well-suited as the gate material of the future, and TiO2, with its high dielectric constant (Table 1), would be an attractive candidate for this application. A new kind of gate oxide must meet verystringent requirementsÐno surface states, virtuallypin-hole free, stoichiometric ultrathin ®lms, good interface formation with the Si substrate, etc. [53].TiO2 could be a viable approach to dielectrics whose oxide equivalent thickness is less than 2.0 nm. CVD-grown TiO2 ®lms on Si show excellent electric characteristics, but a low resistivitylayer, probablySiO 2, forms at the interface [54]. Interestingly, modi®ed TiO2 ®lms are also promising materials for spintronics applications, although TiO2 itself is not a magnetic material. When anatase TiO2 ®lms are doped with a few percent of Co, theybecome ferromagnetic [55,56]. Such ®lms are opticallytransparent, semiconducting, and ferromagnetic at room temperature, and might be ideal candidates for spin-based electronic devices.

Nanostructured TiO2 electrodes have received quite a bit of attention. One particularlyinteresting application is the implementation of nanocrystalline TiO2 ®lms in electrochromic devices [57]. Such devices control light transmission in windows or light re¯ection in mirrors and displays. They are based U. Diebold / Surface Science Reports 48 (2003) 53±229 61

Table 1 Bulk properties of titanium dioxidea Atomic radius (nm) O 0.066 (covalent) Ti 0.146 (metallic)

Ionic radius (nm) O(À2) 0.14 Ti(4) 0.064 Crystal structure System Space group Lattice constants (nm) abcc/a

14 rutile Tetragonal D4h-P42/mnm 0.4584 ± 0.2953 0.644 19 anatase Tetragonal D4h-I41/amd 0.3733 ± 0.937 2.51 15 brookite Rhombohedral D2h-Pbca 0.5436 0.9166 0.5135 0.944

Density(kg/m 3) rutile 4240 anatase 3830 brookite 4170

Melting point (8C) Boiling point (8C) (decomposes) (at pressure

(rutile) pO2 101.325 kPa) 1870 2927

0 Standard heat capacity, Cp, 298.15 J/(mol 8C) 55.06 (rutile) 55.52 (anatase)

Heat capacity, Temperature (K)

Cp (J/kg K) (rutile) ±10 243 25 1788 50 6473 100 10718 150 14026 200 18255 298.15

Temperature (K) Thermal conductivity (W/(m K)) (rutile) 373 6.531 473 4.995 673 3.915 873 3.617 1073 3.391 1273 3.307 1473 3.307 62 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 1 (Continued ) Linear coefficient of Temperature (8C) thermal expansion (a Â 10À6, 8CÀ1), rutile 8.19 0±500

Anisotropyof linear coefficient of thermal expansion (a Â 10À6, 8CÀ1), rutile Parallel to c-axis Perpendicular to Temperature c-axis (8C) a 8:816 Â 10À6 a 7:249 Â 10À6 30±650 3:653 Â 10À9 Â T 2:198 Â 10À9 Â T 6:329 Â 10À12 Â T2 1:198 Â 10À12 Â T2

Modulus of normal Density(kg/m 3) elasticity E (GPa) (rutile) 244.0 4000 254.5 4100 273.0 4200 284.2 4250 289.4

Hardness on mineralogical scale (Mohs scale) 5±6.5 Microhardness (MPa) Load P Â 10À5 N 6001.88 98070 rutile 7845.66±1961.40 49035±98070 rutile, 398±923 K Compressibilitycoefficient, Pressure, p, Temperature b,10À11 m2/N, rutile 1011 m2 (N Pa) (K) 0.59 125 273

Electrical resistance (rutile) Temperature (K) Resistivity( O m) 773 3 Â 105 1073 1:2 Â 102 1473 8:50 Â 102

Thermoelectric properties (rutile) Temperature (K) Thermo-emf coefficient (mV/K) 400 0.75 600 À2.75 800 À6 1000 À9 1200 À12 U. Diebold / Surface Science Reports 48 (2003) 53±229 63

Table 1 (Continued ) Galvanometric properties (rutile) Hall constant (m3/c) (rutile) Temperature (K) 2 Â 10À6 500

Electron mobility, m (cm2/V s) Ã TiO2 (rutile) 1 [209] ÃÃ TiO2 (anatase) 10 [209] Dielectric properties Frequency(Hz) Temperature (K) Dielectric constant rutile, perpendicular 108 290±295 86 to optical axis rutile, parallel to ± 290±295 170 optical axis rutile, perpendicular 104 298 160 to c-axis rutile, along c-axis 107 303 100 Band gap (eV) rutile 3.0 (indirect) [209] anatase 3.2 (indirect) [209]

Refractive index

ng nm np a-TiO2 (rutile) 2.908 ± 2.621 b-TiO2 (anatase) 2.488 ± 2.561 g-TiO2 (brookite) 2.7004 2.5843 2.5831 TiO2 (rutile) 2.9467 ± 2.6506 TiO2 (anatase) 2.5688 ± 2.6584 TiO2 (brookite) 2.809 ± 2.677 Integral normal emissivity, Temperature (K)

eÆ (smooth surface) (rutile) 0.82 400 0.83 500 0.84 600 0.85 700 0.86 800 0.87 900 0.875 1000 0.88 1100 0.89 1200 0.90 1300 Monochromatic normal Wavelength, Temperature

emissivity, eln (powder) (rutile) l (nm) (K) 0.27 1.0 1223 0.15 2.0 1223 64 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 1 (Continued ) 0.20 3.0 1223 0.30 4.0 1223 0.32 5.0 1223 0.50 6.0 1223 0.67 7.0 1223 0.76 8.0 1223 0.80 9.0 1223 0.84 10.0 1223 0.85 11.0 1223 0.86 12.0 1223 0.87 13.0 1223 0.88 14.0 1223 0.89 15.0 1223 Refractive index, n,of rutile single crystal for ordinary(1) and extraordinary(2) rays in visible and IR regions of spectrum, at 298 K a Data from [65] unless noted otherwise. A more detailed compendium of bulk properties of rutile is given in [49].

on two complementaryelectrodes (TiO 2 and WO3 in the case of [57]), which change their color upon reduction/oxidation cycles induced by an electrical current.

Polycrystalline ZnO, TiO2 and SnO2, exhibit a high non-linearitybetween the current densityand the electric ®eld and are thus suitable as `varistors' for the suppression of high transient voltages [58].

Doped TiO2 ceramics have useful varistor properties with non-linearitycoef®cient ( a) values in the range a 3À12, a being de®ned bythe relationship I KVa, where I is the current, V the voltage, and K the proportionalityconstant. The presence of this potential barrier is due to the creation of defects formed during sintering of TiO2 systems. A potential barrier associated with a double space charge distribution can originate at these defects. This phenomenon establishes variable resistance as a function of the applied electric ®eld to the solid. Metallic implants in the human bodyhave a signi®cant economic and clinical impact in the biomaterials ®eld [59]. `Commerciallypure' (CP) titanium (ASTM F67) and `extra-low interstitial' (ELI) Ti±6Al±4V alloy(ASTM F136) are the two most common implant biomaterials. There is an increasing interest in the chemical and physical nature of the oxide layer on the surface of both materials [60]. The oxide provides corrosion resistance and mayalso contribute to the biological performance of Ti at molecular and tissue levels, as suggested in the literature on osseointegrated oral and maxillofacial implants byBranemark, Kasemo and co-workers [61] in Sweden.

1.3. Outline of this review

The geometric structure of various TiO2 surfaces is discussed in Section 2. A detailed knowledge of the surface structure is the crucial ®rst step in obtaining a detailed knowledge of reaction mechanisms U. Diebold / Surface Science Reports 48 (2003) 53±229 65 on the molecular scale. Metal oxide surfaces are prime examples of the close relationship between structure and reactivity [6], as local non-stochiometries or geometric defects directlyaffect the electronic structure. Well-tested models are available for both, `perfect' surfaces as well as surface defects on TiO2. Titanium dioxide crystallizes in three crystallographic phases, and the surfaces of the rutile phase have been investigated extensively. Surface science research on the technologically quite important anatase phase is just starting. The structure and stabilityof metal oxide surfaces can be predicted using the concept of autocompensation [5] or non-polarity [62]. Bulk-truncated models of various rutile and anatase TiO2 surfaces are derived using this concept, and are compared with ab initio calculations and experimental results on surface geometrical models and relaxations. Recent scanning probe microscopyresults have given enormous insight into defect structures at TiO 2 surfaces, and have provided some surprises as well.

Section 3 gives a brief summaryof the electronic structure of TiO 2. Most of the basic understanding of the electronic structure of TiO2 surfaces has been discussed in previous reviews [1], hence this section is kept short. Surface defects that are related to oxygen de®ciencies can be identi®ed with most electron spectroscopies, some of which are discussed in this section.

The growth of metal and metal oxide overlayers on TiO2 substrates is reviewed in Section 4. This is a veryactive and exciting area of research, and almost all metals across the periodic table have been investigated on TiO2. Most of the current literature on metal/TiO2 growth has been summarized in Table 6. It is comforting to see that the basic trends for the propensityof interfacial reactions, growth morphology, geometric structure, and thermal stability that have been identi®ed early on [63] are in agreement with the more recent results.

The surface chemistryof TiO 2 is reviewed in Section 5. The adsorption of inorganic molecules is discussed ®rst, and the results for each group of molecules is summarized in tables. Results on small organic molecules is then reviewed. This section closes with a brief summaryof photoinduced reactions on TiO2 surfaces. A summaryand outlook is given at the end.

2. The structure of TiO2 surfaces

Unraveling the relationship between atomic surface structure and other physical and chemical properties is probablyone of the most important achievements of surface science. Because of the mixed ionic and covalent bonding in metal oxide systems, the surface structure has an even stronger in¯uence on local surface chemistryas compared to metals or elemental semiconductors [6]. A great amount of work has been performed on TiO2 over the years, and has lead to an understanding that is unprecedented for a metal oxide surface. This section starts with a brief description of the bulk structure of titanium dioxide crystals, and their stable crystal planes. Because bulk non-stoichiometries in¯uence the surface properties of TiO2 in a varietyof ways,a short discussion of bulk defects is included as well. A substantial part of the section is devoted to the rutile (1 1 0) surface. The (bulk-truncated) (1 Â 1) surface is known with a veryhigh accuracyfrom experimental as well as theoretical studies. Nevertheless, there are some puzzling disagreements between theoryand experiment in some aspects [64]. Surface defects are categorized in step edges, oxygen vacancies, line defects (closely related to the (1 Â 2) reconstruction), common impurities, and the manifestation of crystallographic shear planes (CSPs) at surfaces. The long-standing argument of the structure of the (1 Â 2) phase seems to be settled, as discussed in Section 2.2.2. STM 66 U. Diebold / Surface Science Reports 48 (2003) 53±229 and, more recently, atomic force microscopy (AFM), studies have revealed the complexity of the seeminglysimple rutile (1 1 0) surface, hence the section on TiO 2(1 1 0) commences with a recommendation on the best wayto prepare this surface. The two other low-index planes, rutile (1 0 0) and (0 0 1) are described in Sections 2.3 and 2.4, respectively. Until fairly recently the (1Â 3) reconstruction of the rutile (1 0 0) seemed well understood, but inconsistencies in theoretical calculations as well as new interpretations of X-raydiffraction data show that a closer look on the structure of this phase maybe needed ( Section 2.3.2.2). New developments on structural investigations of anatase samples are included at the end.

2.1. Bulk structure

14 Titanium dioxide crystallizes in three major different structures; rutile (tetragonal, D4h-P42/mnm, Ê Ê 19 Ê Ê a b 4:584 A, c 2:953 A [49]), anatase (tetragonal, D4h-I41/amd, a b 3:782 A, c 9:502 A) 15 Ê Ê Ê and brookite (rhombohedrical, D2h-Pbca, a 5:436 A, b 9:166 A, c 5:135 A) [65]. (Other structures exist as well, for example, cotunnite TiO2 has been synthesized at high pressures and is one of the hardest polycrystalline materials known [66].) However, onlyrutile and anatase playanyrole in the applications of TiO2 and are of anyinterest here as theyhave been studied with surface science techniques. Their unit cells are shown in Fig. 2. In both structures, the basic building block consists of a titanium atom surrounded bysix oxygenatoms in a more or less distorted octahedral con®guration. In each structure, the two bonds between the titanium and the oxygen atoms at the aspices of the octahedron are slightlylonger. A sizable deviation from a 90 8 bond angle is observed in anatase. In rutile, neighboring octahedra share one corner along h110iÐtype directions, and are stacked with their long axis alternating by90 8 (see Fig. 2 as well as Fig. 6). In anatase the corner-sharing octahedra form (0 0 1) planes. Theyare connected with their edges with the plane of octahedra below. In all three

TiO2 structures, the stacking of the octahedra results in threefold coordinated oxygen atoms. Rutile TiO2 single crystals are widely available. They can be bought in cut and polished form from companies such as Commercial Crystal Laboratories, USA; Kelpin Kristallhandel, Germany; Goodfellow, UK; Earth Jewelry, Japan and many others. A very small roughness is achieved by grinding the sample, and then polishing the surface for manyhours with a chemo-mechanical treatment. This is also referred to as epitaxial polish. Practical aspects of surface preparation and handling are discussed in [67].

Ramamoorthyand Vanderbilt [68] calculated the total energyof periodic TiO 2 slabs using a self- consistent ab initio method. The (1 1 0) surface has the lowest surface energy, and the (0 0 1) surface the highest. This is also expected from considerations of surface stability, based on electrostatic and dangling-bonds arguments discussed in Section 2.2.1.1. below. The thermodynamic stability of the (1 0 0) surface was also considered, and was found to be stable with respect to forming (1 1 0) facets. The (0 0 1) surface was almost unstable with respect to the formation of macroscopic (1 Â 1) (0 1 1) facets. From the calculated energies a three-dimensional (3D) Wulff plot was constructed, see Fig. 3. The Wulff construction [69] gives the equilibrium crystal shape of a macroscopic crystal. For comparison with experimental crystal shapes one has to take into account that only four planes were considered and that the calculations are strictlyvalid onlyat zero temperature. The experimental results on the three low-index rutile surfaces discussed below ®t rather well with the stabilityexpected from these calculations. For rutile, the (1 1 0), (0 0 1) and (1 0 0) surfaces have been studied, with (1 1 0) being the most stable one. These three surfaces are discussed in this section. U. Diebold / Surface Science Reports 48 (2003) 53±229 67

Fig. 2. Bulk structures of rutile and anatase. The tetragonal bulk unit cell of rutile has the dimensions, a b 4:587 AÊ , c 2:953 AÊ , and the one of anatase a b 3:782 AÊ , c 9:502 AÊ . In both structures, slightlydistorted octahedra are the basic building units. The bond lengths and angles of the octahedrallycoordinated Ti atoms are indicated and the stacking of the octahedra in both structures is shown on the right side.

Fig. 3. The equilibrium shape of a macroscopic TiO2 crystal using the Wulff construction and the calculated surface energies of [68]. Taken from Ramamoorthyand Vanderbilt [68]. # 1994 The American Physical Society. 68 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 4. Phase diagram of the Ti±O system taken from Samsonov [65]. The region Ti2O3±TiO2 contains Ti2O3,Ti3O5,seven discrete phases of the homologous series TinO2nÀ1 (Magneli phases), and TiO2. See [65] for a more detailed description.

The two approaches that are commonlyused to predict the structure and stabilityof oxide surfaces are exempli®ed in detail for the rutile (1 1 0) surface. For anatase, the (1 0 1) and the (1 0 0)/(0 1 0) surface planes are found in powder materials, together with some (0 0 1). The (1 0 1) surface was calculated to have the lowest surface energy, even lower than the rutile (1 1 0) surface [70]. First experimental results on anatase (0 0 1) and (1 0 1) are discussed at the end of this section.

2.1.1. Bulk defects The titanium±oxygen phase diagram is very rich with many stable phases with a variety of crystal structures, see Fig. 4 [65]. Consequently, TiO2 can be reduced easily. Bulk reduction and the resulting color centers are re¯ected in a pronounced color change of TiO2 single crystals from initially transparent to light and, eventually, dark blue, see Fig. 5. These intrinsic defects result in n-type doping and high conductivity, see Table 2. The high conductivitymakes TiO 2 single crystals such a convenient oxide system for experimentalists. As has been pointed out recently [71], bulk defects playa major role in a varietyof surface phenomena where annealing to high temperatures is necessary, e.g. during the encapsulation of Pt [18,20,72], in bulk-assisted reoxidation [73,74], in restructuring and reconstruction processes [75,76], and adsorption of sulfur and other inorganic compounds [77]. The relationship between crystal color, conductivity, bulk defects as characterized by EPR measurements, and surface structure of rutile (1 1 0) has been investigated systematically by Li et al. [71], and the samples reproduced in Fig. 5 have been used in this study. The electric properties in dependence on the bulk defect concentration has been investigated in [78,79]. The bulk structure of reduced TiO2Àx crystals is quite complex with a various types of defects such as doublycharged oxygenvacancies, Ti 3 and Ti4 interstitials, and planar defects such as CSPs. The defect structure varies with oxygen de®ciency which depends on temperature, gas pressure, impurities, etc. Despite years of research, the question of which type of defect is dominant in which region of oxygen de®ciency is still subject to debate [78,80]. It was shown that the dominant type are Ti 18 19 interstitials in the region from TiO1.9996 to TiO1.9999 (from 3:7 Â 10 to 1:3 Â 19 missing O atoms U. Diebold / Surface Science Reports 48 (2003) 53±229 69

Fig. 5. Color centers associated with bulk defects that are formed upon reduction of TiO2 single crystals cause a change in crystal color. (a) Photograph of rutile single crystals heated in a furnace to various temperatures: (cube 1) 19 h at 1273 K, (cube 2) 21 h 40 min at 1450 K (was like cube 3) then reoxidized in air at 1450 K, (cube 3) 4 h 55 min at 1450 K, (cube 4) 35 min at 1450 K, (cube 5) 1 h 10 min at 1350 K. (b) Same samples after prolonged experiments on cubes 1, 3, and 4. The samples were sputtered dailyand annealed to 973 K for a total of 690 min. Adapted from Li and co-workers [71]. # 2000 The American Chemical Society.

per cubic centimeter) [78]. CS planes precipitate on cooling crystals across the TiO2Àx (0 x 0:0035) phase boundary [81]. Theyshow a verystrong dependence on the cooling historyand are absent in quenched specimen. The formation mechanism was reviewed bySmith et al. [81±83].SuchCSplanes mayextend all the wayto the surface [84±88] and their appearance is discussed in Section 2.2.1.4.

Table 2 a Resistivity( O cm) at 300 K measured at room temperature of different TiO2 samples Cube 2 Cube 5 Cube 1 Cube 4 Cube 3 Resistivity1835.0 108.24 46.76 24.06 8.94 a The colors of cubes 1, 3, and 4 are shown in Fig. 5b; cubes 2 and 5 were additionallyreduced. From [156]. 70 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 6. (a) Ball-and-stick model of the rutile crystal structure. It is composed of slightly distorted octahedra, two of which are indicated. Along the [1 1 0] direction these octahedra are stacked with their long axes alternating by90 8. Open channels are visible along the [0 0 1] direction. The dashed lines A and B enclose a charge-neutral repeat unit without a dipole moment perpendicular to the [1 1 0]-direction (a `type 1' crystal plane according to the classi®cation in [62]). (b) The crystal is `cut' along line A. The same number of Ti ! O and O ! Ti bonds are broken, and the surface is autocompensated [5]. The resulting (1 1 0) surfaces are stable and overwhelming experimental evidence for such (1 Â 1)-terminated TiO2(1 1 0) surfaces exists.

The diffusion mechanism for the various types of defects is quite different; oxygen migrates via a site exchange (vacancydiffusion) mechanism, while excess Ti diffuses through the crystalas interstitial atoms. The interstitial diffusion happens especiallyfast through the open channels along the (0 0 1) direction (the crystallographic c-axis) [89,90], see Fig. 6a. A Ti interstitial located in these channels is in an octahedral con®guration, similar to the regular Ti sites [91]. Consequently, the diffusing species in oxidation reactions of reduced TiaOb surfaces (where a > b=2 but probablyless than b) produced by sputtering and/or Ti deposition is the Ti atom and not the O vacancy, as has been shown in a series of elegant experiments with isotopicallylabeled 18O and 46Ti byHenderson [73,74].

2.2. The structure of the rutile TiO2(1 1 0) surface

The rutile (1 1 0) surface is the most stable crystal face and simple guidelines can be used to essentiallypredict the structure and the stabilityof TiO 2(1 1 0)-(1 Â 1). Because these concepts are veryuseful for the other crystalfaces of TiO 2 as well other oxide materials, theyare exempli®ed for this surface. The relaxations from the bulk-terminated coordinates are reviewed, and the types and manifestations of defects are discussed. Although the TiO2(1 1 0) surface is verystable, it nevertheless reconstructs and restructures at high temperatures under both oxidizing and reducing conditions.

2.2.1. The (1 Â 1) surface

2.2.1.1. Bulk truncation. Two concepts have been introduced to predict the stabilityof oxide structures. Tasker [62] discussed the stabilityof ionic surfaces based on purelyelectrostatic considerations. U. Diebold / Surface Science Reports 48 (2003) 53±229 71

The second concept, autocompensation, was originallydeveloped for surfaces of compound semiconductors and applied to metal oxide surfaces byLaFemina [5]. The most stable surfaces are predicted to be those which are autocompensated, which means that excess charge from cation-derived dangling bonds compensates anion-derived dangling bonds. The net result is that the cation- (anion-) derived dangling bonds are completelyempty(full) on stable surfaces. This model allows for the partially covalent character found in manymetal oxides, including TiO 2. Both concepts are used in a complementaryway,and represent a necessary(but not sufficient) condition for stable surface terminations. Veryoften, stable metal oxide surfaces for which the structure is known are non-polar [62] and fulfill the autocompensation criterion [5]. Tasker's and LaFemina's approaches are exempli®ed in creating a stable (1 1 0) surface (Fig. 6). In Tasker's concept, the dipole moment of a repeat unit perpendicular to the surface must be zero in order for the surface energyto converge. He introduced three categories for ionic (or partiallyionic) structures. Type 1 (neutral, with equal number of cations and anions on each plane parallel to the surface) is stable. Type 2 (charged planes, but no dipole moment because of a symmetrical stacking sequence) is stable as well. Type 3 surfaces (charged planes and a dipole moment in the repeat unit perpendicular to the surface) will generallybe unstable. Consider, for example, the rutile structure as being composed of (1 1 0)-oriented planes such as drawn in Fig. 6a. The top plane in Fig. 6a consists of the same number of Ti and O atoms. In a purely ionic picture, the titanium and oxygen atoms have nominal charges of 4 and À2, respectively. Hence, the top layer has a net positive charge. The next two layers consist of oxygen atoms, hence both of them have a net negative charge. A Type 2 repeat unit is outlined by the dashed lines A and B in Fig. 6a. It consists of a mixed Ti, O layer, sandwiched between two layers of oxygen atoms. The total unit does not have a dipole moment (and from counting the charges it turns out that it is neutral as well). A crystal, cut or cleaved1 to expose a (1 1 0) surface, will naturallyterminate with the surface created bycutting along line A (or B) in Fig. 6a. In Fig. 6b, the top of the model is shifted along the (1 1 0) direction (cutting the crystal in a `Gedankenexperiment'). The resulting surface is very corrugated because one `layer' of oxygen atoms is left behind. As shown below, there is overwhelming evidence that the (1 Â 1) surface of

TiO2(1 1 0) closelyresembles the `bulk-terminated' structure depicted in Fig. 6b. The same surface structure is also predicted using the rules of autocompensation. In Fig. 6b, the same number of oxygen-to-titanium bonds are broken as titanium-to-oxygen. Transferring electrons from the dangling bonds on the Ti cations will just compensate the missing charge in the dangling bonds on the O anions. Hence, the surface is autocompensated [5]. Note that onlythe longer bonds are broken when the crystal is sliced in this way. The rutile (1 1 0)-(1 Â 1) surface in Fig. 6b contains two different kinds of titanium atoms. Along the [0 0 1] direction, rows of sixfold coordinated Ti atoms (as in the bulk) alternate with ®vefold coordinated Ti atoms with one dangling bond perpendicular to the surface. Two kinds of oxygen atoms are created as well. Within the main surface plane, theyare threefold coordinated as in the bulk. The so- called bridging oxygen atoms miss one bond to the Ti atom in the removed layer and are twofold coordinated. These bridging oxygen atoms are subject to much debate. Because of their coordinative undersaturation, atoms from these rows are thought to be removed relativelyeasilybythermal annealing. The resulting point defects (Section 2.2.1.4) affect the overall chemistryof the surface, even in a macroscopic way [92].

1 Unfortunately, TiO2 fractures and does not cleave well. 72 U. Diebold / Surface Science Reports 48 (2003) 53±229

A(1Â 1) LEED pattern is generallyobserved upon sputtering and annealing in UHV. To this author's knowledge no quantitative LEED studyhas been reported, probablybecause of the defects are easilycreated when the sample is bombarded with electrons which poses an additional complication

(see Section 2.2.1.4). A medium-energyelectron diffraction (MEED) studyof TiO 2(1 1 0) employed an ESDIAD optics with a channelplate; this setup is more sensitive than a conventional LEED apparatus, and allows for verysmall electron currents to be used. The results of this studywere consistent with the (1 Â 1) structure depicted in Fig. 6b. X-rayphotoelectron diffraction (XPD) spectra also ®t the expected (1 Â 1) termination [93], as do the STM results discussed in Section 2.2.1.3.

2.2.1.2. Relaxations. Everysurface relaxes to some extent. In recent years,the geometryof the

TiO2(1 1 0)-(1 Â 1) surface has been studied in some detail both experimentallyand theoretically. The results of a surface X-raydiffraction (SXRD) experiment [94] and of several total-energy calculations are listed in Table 3. The experimentallydetermined directions of atoms in the first layers are sketched in Fig. 7. As is expected from symmetry, the main relaxations occur perpendicular to the surface. Onlythe in-plane oxygens(4, 5 in Fig. 7) move laterallytowards the fivefold coordinated Ti atoms. (These relaxations are symmetric with respect to the row of fivefold coordinated Ti atoms, hence do not increase the size of the surface unit cell.) The bridging oxygen atoms (labeled 3 in Fig. 7) are measured to relax downwards considerably, and the sixfold coordinated Ti (1) atoms upwards. The fivefold coordinated Ti atoms (2) move downwards and the neighboring threefold coordinated oxygen

Table 3 Displacements (AÊ ) determined experimentallyand theoreticallybyseveral groups using different computational techniques a Charlton, Harrison, Harrison, Rama-moorthy, Bates, Lindan, Vogten-huber, Reinhardt, SXRD, FP-LAPW, LCAO, PW-PP-LDA, PW-GGA, PW-PP-GGA, FP-LAPW, HF-LCAO, experiment seven layers seven layers five layers five layers three layers three layers three layers Ti(1) (sixfold) 0.12 Æ 0.05 0.08 0.23 0.13 0.23 0.09 À0.05 0.09 Ti(2) (®vefold) À0.16 Æ 0.05 À0.23 À0.17 À0.17 À0.11 À0.12 À0.18 À0.15 O(3) (bridging) À0.27 Æ 0.08 À0.16 À0.02 À0.06 À0.02 À0.09 À0.16 À0.14 O(4,5) [1 1 0] 0.05 Æ 0.05 0.09 0.03 0.12 0.18 0.11 À0.12 0.07 1 10Æ0.16 Æ 0.08 Æ0.06 Æ0.05 Æ0.04 Æ0.05 Æ0.05 Æ0.07 Æ0.08 O(6) 0.03 Æ 0.08 À0.09 0.02 À0.07 0.03 À0.05 ± À0.07 Ti(7) 0.07 Æ 0.04 0.07 0.14 0.06 0.12 ± ± ± Ti(8) À0.09 Æ 0.04 À0.13 À0.10 À0.08 À0.06 ± ± ± O(9) 0.00 Æ 0.08 À0.05 0.00 0.02 0.03 ± ± À0.02 O(10,11) [1 1 0] 0.02 Æ 0.06 À0.04 0.03 À0.03 0.00 ± ± ± Æ0.07 Æ 0.06 Æ0.03 Æ0.03 Æ0.05 Æ0.02 O(12) À0.09 Æ 0.08 À0.04 À0.01 À0.01 0.03 ± ± ± Ti(13) ± 0.02 0.05 ± ± ± ± ± Ti(14) ± À0.08 À0.06 ± ± ± ± ± O(15) À0.12 Æ 0.07 À0.07 0.01 ± ± ± ± ± O(16,17) [1 1 0] ± À0.03 0.01 ± ± ± ± ± [1±10] ± Æ0.02 Æ0.02 O(18) ± À0.02 0.01 ± ± ± ± ± O(19) ± 0.02 À0.01 ± ± ± ± ± a The atomic labels and the directions of the experimentallydetermined relaxations are given in Fig. 7. The results are grouped by®rst authors: Charlton [94], Harrison [64], Ramamoorthy [68], Bates [249], Lindan [233], Vogtenhuber [100], and Reinhard [101]. Acronyms used are SXRD (surface X-raydiffraction), FP-LAPW (full-potential linear augmented plane wave), LCAO (linear combination of atomic orbitals), HF (Hartree±Fock), PW-PP (plane-wave pseudopotential), LDA (local densityapproximation), and GGA (generalized gradient approximation).

Indicated are the number of TiO2 repeat units used for the various calculations. Expanded from a similar compendium given in [64]. U. Diebold / Surface Science Reports 48 (2003) 53±229 73

Fig. 7. Model of the TiO2(1 1 0)-(1 Â 1) surface. The relaxations of surface atoms, determined with SRXD are indicated [94]. The labels refer to the relaxations listed in Table 3. Redrawn from Charlton et al. [94]. # 1997 The American Physical Society.

atoms (4, 5) upwards, causing a rumpled appearance of the surface. The relaxations in the second TiO2 layer are approximately a factor of two smaller. The most striking feature in the experimentallydetermined (relaxed) coordinates is the large relaxation of the bridging oxygen atoms by À0.27 AÊ . The measured geometrywould indicate a very small bond length between the sixfold coordinated Ti atom (1) and the bridging oxygens (3) of only 1:71 Æ 0:07 AÊ instead of the 1.95 AÊ expected from the bulk structure. The relaxation results in vertical distances of 0:89 Æ 0:13 and 1:16 Æ 0:05 AÊ from the sixfold (1) and ®vefold coordinated (2) Ti atoms, respectively. This is in agreement with ion scattering measurements, where vertical distances of 87 and 1:05 Æ 0:05 AÊ were found [95,96]. (Another ion scattering studyfound the height of the bridging oxygen atoms comparative to that of the bulk structure but the interlayer distance largely relaxed with about À18 Æ 4% [97].) Photoelectron diffraction results [98] are also in agreement with relaxations from the X-raydiffraction work given in Table 3. The results of total-energycalculations byseveral groups [64,68,99±102] are compared to the measured relaxations in Table 3. Two complementaryapproaches were used, the linear combination of atomic orbitals (LCAOs) and plane-wave techniques. Either periodic or free-standing supercells with different numbers of layers (in the sense of Tasker's non-polar repeat units in Fig. 6a) were used. For example, the con®guration drawn in Fig. 7 represents part of the upper half of the seven-layer slab used byHarrison et al. [64]. Because of the localized nature of the Ti3d electrons in the TiO2 structure, plane-wave expansions are challenging. A rather high-energycutoff needs to be used for convergence, and the functional for the LDA- or GGA-based calculations mayalso in¯uence the results [103].In addition, the thickness of the slab mayplaya role in the accuracyof the calculated geometry. 74 U. Diebold / Surface Science Reports 48 (2003) 53±229

The directions of the calculated relaxations agree in (almost) all the theoretical papers with the experimentallydetermined coordinates. The quantitative agreement is not as good as one could expect from state-of-the art ab initio calculations, however. As Harrison et al. [64] pointed out, the extensive experience of calculations on bulk oxides which has been built up in recent years leads one to expect that DFT and HF calculations will reproduce experimental bond lengths to somewhat better than 0.1 AÊ . Ê For example, the bulk structural parameters of TiO2 rutile agree better than 0.06 A using soft-core ab initio pseudopotentials constructed within the LDA, and a plane-wave basis [104]. In particular, all the calculations ®nd a much smaller relaxation for the position of the bridging oxygen atom. A possible reason for this disagreement was given by Harrison et al. [64]. All the theoretical results listed in Table 3 are strictlyvalid onlyat zero temperature. It is conceivable that strong anharmonic thermal vibrations at the TiO2(1 1 0) surface cause the discrepancybetween experimental and theoretical results. However, molecular dynamics simulations using the Carr± Parinello approach [105] found that the average position in dynamic calculations is only relaxed by 0.05 AÊ rather than by0.27 AÊ , discarding this explanation. Instead, it was suggested that the O atom might relax laterallyso that it is displaced into an asymmetricposition. Based on these theoretical results, the ®nite temperature has to be taken into account for a proper evaluating diffraction results. Hopefully, future experiments will show whether a better agreement with theoreticallypredicted relaxations can be achieved. When considering surface reactions, one also needs to depart from a static picture of this and other oxide surfaces, and has to keep in mind the substantial distortions and bond length changes that take place during such large-amplitude vibrations. It is now well-known that adsorbates often have a signi®cant in¯uence on `re-relaxing' the surface. Computational studies, e.g. the one given in [106] for the adsorption of Cl, clearlyshow strong effects upon adsorption. Onlya few experimental exist so far. For example, Cu overlayerson TiO 2(1 1 0) cause the Ti atoms at the Cu/TiO2(1 1 0) interface relax back to the original, bulk-like positions. The O atoms relax even stronger, which was attributed to Cu±O bonding [107].

2.2.1.3. Appearance in STM and AFM. Naturally, scanning probe techniques are extremely useful tools for studying atomic-scale structures at TiO2 and other metal oxide surfaces, where local changes in stoichiometryor structure can severelyaffect surface reactivity.On TiO 2(1 1 0), STM and, more recently, non-contact AFM, have been used bymanydifferent groups. These techniques have provided valuable and verydetailed insight into local surface structure. However, the interpretation of STM images of oxides is somewhat challenging because of strong variations in the local electronic structure, and because tips can easily`snatch' a surface oxygenatom, which can cause a change in tip states and result in `artifacts' in STM images. There is now consensus among different groups on what is `really' observed with STM, at least under `normal' operating conditions.

The dominant tunneling site on TiO2(1 1 0) surfaces has been subject to some debate in the past. In principle, there is uncertaintyas to whether the image contrast is governed bygeometric or electronic- structure effects. For TiO2, atomic-resolution STM is often onlysuccessful when imaging unoccupied states (positive sample bias) on reduced (n-type) samples. In reduced TiO2 crystals, the Fermi level is close to the conduction-band minimum (CBM) in the 3 eV gap, and electronic conduction occurs predominantlythrough high-lyingdonor states [78]. Under a typical bias of 2 V, electrons can thus tunnel from the tip into states within 2 eV above the CBM, and be conducted awayfrom the surface. On the one hand, these CBM states have primarilycation 3d character (the valence band having primarilyO 2p character, see Section 3) so that one might expect to image the metal atoms as the U. Diebold / Surface Science Reports 48 (2003) 53±229 75

Ê Fig. 8. STM image of a stoichiometric TiO2(1 1 0)-(1 Â 1) surface, 140 Ð Â 140 A. Sample bias 1.6 V, tunneling current 0.38 nA. The inset shows a ball-and-stick model of the unrelaxed TiO2(1 1 0)-(1 Â 1) surface. There is now overwhelming evidence that the contrast on this surface is normallyelectronic rather than topographic, and that the bright lines in STM images normallycorrespond to the position of the Ti atoms rather than the bridging oxygenatoms. From Diebold et al. [116]. # 1998 The American Physical Society.

``white'' features in STM topographs. On the other hand, the bridging oxygen atoms protrude above the main surface plane and dominate the physical topography (see inset in Fig. 8). Hence it seems equally plausible that geometrical considerations might dominate the contrast in STM images. Fig. 8 shows an STM image of a stoichiometric (1 Â 1) surface. Bright and dark rows run along the [0 0 1] direction in Fig. 8. The distance between the rows is 6:3 Æ 0:25 AÊ , in agreement with the unit cell dimension of 6.5 AÊ along 1 10. At neighboring terraces theyare staggered byhalf a unit cell. It is not immediatelyobvious if these bright rows correspond to lines of bridging oxygenatoms or ®vefold coordinated Ti4 ions. The ``bridging oxygen'' rows protrude from the surface plane on a relaxed

TiO2(1 1 0) surface (see Table 3), so if STM were dominated bytopographical effects, theywould appear as rows with high contrast in Fig. 8. There is strong evidence that, normally, this is not the case, and that the Ti sites are imaged bright in this and similar images. Onishi and Iwasawa [108] have observed formate ions (which are expected to adsorb to Ti sites) on top of the bright rows. This is now con®rmed for manyother adsorbates, e.g. chlorine [109] and sulfur [77] appear as bright spots on top of bright or dark rows when adsorbed on Ti sites or oxygen sites, respectively, see Fig. 56 in Section 5.1.6. 76 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 9. Contour plots of [0 0 1]-averaged charge densities associated with electron states within 2 eV of the CBM for (a) the relaxed stoichiometric (1 Â 1) surface, and (b) the relaxed oxygen-de®cient (1 Â 2) surface. Contour levels correspond to a geometric progression of charge density, with a factor of 0.56 separating neighboring contours. To ®rst approximation, the STM tip will follow one of the equal-densitycontours several AÊ ngstroms above the surface. From [110,116].

A theoretical approach to determine the image contrast in STM is shown in Fig. 9. Pseudopotential calculations were used to analyze the local density of states in the vacuum region above the surface [110]. In rough correspondence with the experimental bias conditions, the charge densityof conduction-band states were summed up from 0 to 2 eV above the conduction-band minimum. This quantitywas then averaged over the [0 0 1] direction and plotted as a function of the other two coordinates as shown in Fig. 9. Under constant-current tunneling conditions, the STM tip is expected to follow roughlyone of the equal-densitycontours several A Ê ngstroms above the surface. The plot in Fig. 9a clearlyshows that the charge-densitycontours extend higher above the ®vefold coordinated Ti atoms when the tip is a few AÊ above the surface, in spite of the physical protrusion of the bridging oxygen atoms. This con®rms that the STM is imaging the surface Ti atoms, i.e., that the apparent corrugation is reversed from the physical one by electronic-structure effects. The slab in Fig. 9b has a (1 Â 2) symmetry with every other bridging oxygen row missing. This con®guration has been proposed originallyto account for the (1 Â 2) structure observed in LEED [111]. More recent experimental evidence, resulting predominantlyfrom STM measurements, has shown that this is not a likelystructure (see Section 2.2.2). However, the charge densitycontours in Fig. 9b indicate that single vacancies in the bridging oxygen rows are expected to appear as bright features on the dark oxygen rows. A different and computationallyless expensive computational approach has been taken byGu Èlseren et al. [112]. Theyhave used a ®rst-principles atomic-orbital base scheme with limited self-consistency. From analyzing the radial distributions of O and Ti wave functions, it was concluded that the STM tip should sample electrons from different surface atoms, depending on the tip±sample separation. For close distances (<4 AÊ ) the contributions from the oxygen atoms should dominate, while for larger separations, the Ti atoms are dominant. Hence a `reversal' of the tunneling site should be possible. STM images, taken with high tunneling current and relativelylow bias voltages ( It 2:0 nA, Vs 0:75 V) seem to con®rm this conclusion [113]. Such images show an enhanced resolution, and U. Diebold / Surface Science Reports 48 (2003) 53±229 77

Fig. 10. Two examples for spontaneous tip changes that give rise to a changed appearance of STM images of TiO2(1 1 0)- (1 Â 1). When the tip is treated with high voltage/high current pulses the `normal' tip state is usuallyre-gained that renders images as shown in Fig. 8. From [116]. # 1998 Elsevier. alternating rows of individual dots and white rows. High resolution was also reported in [114] and has been explained as tunneling centered at the ®vefold coordinate Ti4 ions and at the bridging oxygen ions. After `functionalizing' the STM tip byscanning over a Si surface with 10 V sample bias and 20 nA, enhanced resolution has been observed byanother group [115]. These images were interpreted as tunneling into the sixfold coordinated Ti atoms underneath the bridging oxygen's. The tunneling process was interpreted as being in¯uenced bya strong chemical interaction and formation of a partial chemical bond between Si at the tip and surface oxygen. Two examples for spontaneous tip changes are shown in Fig. 10 [116]. In the upper half of Fig. 10a, the lateral resolution appears to be enhanced as compared to `normal' images, and additional small, bright spots are visible between the bright rows in some areas. This change in appearance did not result from an intentional lowering of the tunneling resistance as in [113], but was interpreted as an interference effect caused bya double tip with widelyspaced apex atoms that are tunneling on different terraces bythe authors in [116].InFig. 10b, a spontaneous tip change occurred about half- waythrough the image. At the lower half, bright spots are located between the bright rows. As indicated in the context of Fig. 9b, the position and appearance of oxygen vacancies can be taken as a tell-tale signal on whether the row of O or Ti are imaged, and the lower half is consistent with imaging the Ti rows. In one report, dark spots on bright rows have been assigned as point defects at

TiO2(1 1 0) surfaces [117]. Consequently, bright rows were assigned as the location of bridging oxygen atoms. The images shown in [117] resemble the upper part of Fig. 10b, where pronounced black spots appear on the bright rows. At ®rst sight, the image shown in Fig. 10b could be taken as an indication for image reversal caused bysuch a compositional change. Note that the bright rows continue as dark rows on the right side of Fig. 10b. However, on the left side of Fig. 10b the bright rows are in phase on the upper and lower part of the image. This indicates that the observed change in contrast is caused bya lateral shift of the outermost atom on the tip apex rather than byan actual image reversal. 78 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 11. Non-contact AFM image of a TiO2(1 1 0)-(1 Â 1) surface. From Fukui et al. [118]. # The American Physical Society.

A de®nite interpretation of images as shown in Fig. 10 is rather dif®cult. For the sake of studying surface structure and adsorbates, it is maybe better (and suf®cient) to focus on results obtained with the `normal' tip state that can be obtained reproduciblyby`cleaning' with high voltage/high current pulses. As mentioned above, onlyfew reports exist where satisfactoryimages have been obtained with negative sample bias (®lled-state images) [77]. These were taken with a bias voltage that is too small to bridge the 3 eV gap. It is likelythat a real `contrast reversal' would occur under the tunneling conditions where the ®lled state of the VB are imaged, but, to this author's knowledge, such images have not yet been reported.

Recently, non-contact AFM has been introduced as a complementary technique to study TiO2 surfaces with atomic resolution. An image of the TiO2(1 1 0)-(1 Â 1) surface, obtained byFukui et al. [118], is reproduced in Fig. 11. Frequencymodulation described in [119] was used as the feed back signal. While the contrast formation of atomicallyresolved AFM images using this technique is also somewhat controversial [120], one would assume that the physical geometry should dominate in AFM. The registryof (1 Â 2) strands (see Section 2.2.2) in AFM images is consistent with the bright rows in Fig. 11 being the protruding bridging oxygens. Consequently, the black spots in Fig. 11 were assigned as oxygen vacancies in [118].

2.2.1.4. Surface defects. The abilityto control the amount of defects is one of the main attractions of TiO 2 as a `well-characterized' model system. Because imperfections such as vacancies introduce changes in the electronic structure (in particular a band gap feature at 0.8 eV below EFermi, and a shoulder in the XPS Ti2p peak, see Section 3.2), theyhave been investigated with spectroscopic techniques for years. Much has been learned about the structure of defects, mainlybecause of recent investigations with scanning probe techniques. The following discussion considers steps; vacancies produced bythermal annealing, sputtering, and electron bombardment; as well as common impurities such as Ca and H.

2.2.1.4.1. Step edges. Sputtering and annealing in UHV (at not too high temperatures) renders flat (1 Â 1) surfaces. As is expected for annealing of sputter-damaged surfaces, the terrace size increases with U. Diebold / Surface Science Reports 48 (2003) 53±229 79

Fig. 12. STM image of a clean stoichiometric TiO2(1 1 0)-(1 Â 1) surface after sputtering and annealing to 1100 K in UHV. The step structure is dominated bystep edges running parallel to h1 11i and h001i directions. A kink site at a h1 11i step edge is marked with `K'. Smooth (`UR') and rugged (`R'econstructed) h001i-type step edges appear with roughly equal probabilityand are marked with arrows. The inset shows a 100 Ð Â 100 Ð wide image of a reconstructed step edge. From [116]. annealing temperatures. This has been shown nicelyin an STM work byFischer et al. [117]. The correlation length in SPA-LEED measurements (which corresponds to the average terrace size) has increases with a T1/4 dependence at temperatures above 800 K [121]. Interestingly, Ar implanted during the sputtering process at relativelymoderate ion energies (1000 eV) and fluences (typically1 mA/cm2 and 30 min) does not completelyleave the near-surface region during annealing up to 1000 K and is still visible in XPS and AES [77]. An example for a typical terrace-step structure is shown in Fig. 12. Step edges on annealed surfaces run predominantlyparallel to h001i- and h1 11i-type directions [116,117,122]. These steps are measured to be 3.2 AÊ high, in agreement with the value expected from the rutile structure [123].In Fig. 12, a kink site at the point where a h1 11i-type step edge turns into a h1 1 1i-type step edge is labeled with K. Such kink sites are located at the end of dark rows (the bridging oxygens). Two kind of h001i-type steps are pointed out by arrows in Fig. 13. One kind appears smooth (marked as UR), the other one rugged (marked as R for reconstructed) with a high number of kinks. The inset in the upper left hand corner shows a blow-up (100 Ð Â 100 AÊ ) of a rugged step edge. Both types of step edges appear with roughlyequal probabilityin images. It is relativelystraightforward to construct models for step edges following the rules of autocompensation [5]. For example, Fig. 13 shows a ball-and-stick model of two layers of TiO2(1 1 0) 80 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 13. Ball-and-stick model of two terraces of TiO2(1 1 0). Small black balls represent Ti atoms and large white balls represent oxygen atoms. The step edge FG runs parallel to h1 11i-type directions. The smooth and rugged step edges along h001i in Fig. 12 are attributed to step edges AB and HI, respectively. From [116]. that contains several step edges [116]. As outlined above (Section 2.2.1.1) the same number of O ! Ti bonds and Ti ! O bonds need to be broken when cutting a TiO2 crystal, for example, by forming a step edge byremoving part of the upper terrace. A h1 11i step edge (parallel to the diagonal of the surface unit cell) runs between the corners labeled F and G. For clarity, the bonds along this step edge are shaded with graycolor. The orientation of the step plane is 1 15. The O atoms along the h1 11i step edge in Fig. 13 are alternatelythreefold (as in the bulk) and twofold coordinated. Formation of the step edge creates fourfold coordinated Ti atoms (terminating the Ti rows of the upper terrace) and ®vefold coordinated Ti atoms (terminating the bridging O rows). It should be pointed out that either these fourfold coordinated Ti atoms have a formal oxidation state of 4, or their concentration is verylow, because photoemission results of `unreduced' surfaces show no evidence of Ti3. A change in step orientation from the h1 11i to the h1 1 1i direction occurs always at the bridging oxygen rows (K in Fig. 12). The local environment of the atoms at such a kink site is not different from the rest of the step edge. Similarlyconstructed models of step edges oriented along 1 12 and 1 15 directions are given in [124]. There are several possibilities to form step edges parallel to the h001i direction. One can either cut next to the Ti atoms underneath the bridging oxygens (parallel to the arrow labeled 1 on the upper terrace in Fig. 13) or between the in-plane oxygen and titanium atoms (parallel to arrow 2). If one cuts at position 1, an autocompensated step edge is formed: for each Ti ! O bond that is broken in the upper plane, one O ! Ti bond is broken between the newlyformed bridging oxygensof the lower plane and the ®vefold coordinated titaniums of the upper plane. Note that there are two different terminations for such a step edge; the terrace maybe terminated either bya row containing in-plane oxygen atoms (step edge DC in Fig. 13) or bya row of bridging oxygens(step edge AB). Because of the observed contrast at step edges (smooth step edges terminate with a dark row, Fig. 12), a termination with bridging oxygen atoms (step AB in Fig. 13) is favored. (Fischer et al. [117] have presented a model for a h001i step terminating in in-plane oxygen rows. However, these authors adapted a different interpretation of the bright rows in STM as being caused bybridging oxygenrows.) If one cuts a terrace parallel to the h001i direction at position 2, the step edge that is formed is not autocompensated: onlyoxygen ! titanium bonds are broken on both the upper and on the lower U. Diebold / Surface Science Reports 48 (2003) 53±229 81 terrace. Hence, a step edge that terminates in a row of in-plane Ti atoms is not stable and may reconstruct. This is consistent with the observation of reconstructed step edges (R in Fig. 13) appearing whenever a terrace would terminate in a bright row. A `reconstruction' was proposed [116] byremoving three of four Ti atoms as well as the neighboring in-plane O atoms (step edge HI in Fig. 13). Such a structure is consistent with the bright bumps separated by12 AÊ that are observed in STM. It also ful®lls the criterion of autocompensation. As has been point out before [116], step edges that run parallel to h1 10i type directions (step edge BC in Fig. 13) are generallynot observed. This also ®ts well into the concept of autocompensation. Such a step edge would not be not autocompensated and therefore is not expected to be stable. It should be noted that step edges playan important role in the 1 Â 1! 1 Â 2 phase transformation [125]. It is interesting that the two-step h001i terminations with verydifferent geometries are seen. So far no temperature-dependent STM studies have been performed to ®nd out how annealing temperature affects the relative contribution of these two step edge terminations. In principle, the presence of verysmall amounts of trace impurities can also not be ruled out. In this sense, the model in Fig. 13 for the reconstructed step edge must remain speculative until supported bymore theoretical or experimental evidence. This detailed insight into step geometries and coordination number of atoms at step edges and kink sites is important, because a decrease in coordination number of surface atoms often correlates with an enhancement in chemical reactivity. Microscopic or nanoscopic particles naturally exhibit a much higher step/kink concentration than ¯at single crystals used for surface-science studies, so this issue is even more important in applications that use such materials, see Section 1.2. A few systematic studies of the effect of step edges on surface chemistryunder UHV conditions have been reported. For example, pyridine molecules have been found to be more strongly adsorbed at fourfold coordinated Ti atoms at step sites than at the ®vefold coordinated Ti atoms on the terraces [124]. On the other hand, the opposite effect has been observed byIwasawa et al. [126] for adsorption of formic acid on TiO2(1 1 0). Adsorption at 400±450 K resulted in particles with a strongly suppressed presence in the vicinityof step edges. Possibly, electrostatics plays a role. To this author's knowledge, virtually no theoretical work has been done to determine step geometrywith the same detail as ¯at surfaces. Because step edges break the symmetry, ®rst-principles calculations would require huge unit cells. With the advent of ever faster computers and more powerful programs such calculations maysoon become viable.

2.2.1.4.2. Oxygen vacancies created by annealing. There is overwhelming spectroscopic and chemical evidence for the presence of point defects on samples sputtered and annealed in UHV. These are attributed to vacancies in the bridging oxygen rows. Their concentration is typically reported as several percent [127,128]. These defects are of high importance for the surface properties of TiO2(1 1 0). No systematic study on the correlation between defect concentration and bulk properties has been performed on single crystals. EPR studies on a polycrystalline TiO2 powder [129], reduced at temperatures between 723 and 923 K in vacuo, showed that the ratio of surface-to-bulk Ti3 cations decreases as the reduction temperature is increased. Thermallyinduced point defects are visible as distributed black points on the bright oxygenrows in the AFM image in Fig. 11. As alreadypointed out, STM results are harder to interpret because of strong electronic effects. STM images of titanium dioxide surfaces that have been annealed in UHV often exhibit two kinds of atomic-scale features (Fig. 14). These appear as short bright features centered on dark rows (labeled A) and are connecting neighboring bright rows, and as dark spots on the bright rows (labeled B). The features labeled A have a densityof 7 Æ 3% per surface unit cell, consistent with O 82 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 14. STM image 200 Ð Â 200 Ð of a TiO2(1 1 0) surface, sputtered and annealed in UHV to 1100 K for 10 min, showing point defects. Features labeled with À' have been assigned as oxygen vacancies. The position of the line scan shown is indicated in the image. From [116]. vacancydensityestimates from spectroscopic measurements. The size of the bright spots is of the order of one single atom (FWHM 5 AÊ in line scans along the [0 0 1] direction) when taking into account convolution with the tip. Theyalwaysappear as isolated spots with no apparent short-range ordering but a slight tendencyto be staggered perpendicular to the rows. It is well-known that oxygenvacancies are healed bydosing a reduced TiO 2 sample with oxygen [127,130,131]. When the sample in Fig. 14 was stepwise exposed to oxygen, the density of the bright spots (A) decreased. This, and the contrast expected from electronic structure calculations (missing bridging oxygens give rise to a protrusion in empty-states charge-density contours, see Fig. 9b) led to the assumption of the bright spots being missing oxygen vacancies [110,116]. As was reported in [116], their appearance in STM is stronglytip- dependent. The dark features (B in Fig. 14) are less common (surface concentration of 1±2%) than the oxygen vacancies. Theycan extend over several unit cells. Upon adsorption of oxygen,the number of the dark spots stays constant within the statistical error. They were tentatively assigned as subsurface oxygen vacancies [109], however, ®rst-principles total-energycalculations show that such a con®guration is highlyunlikely [132]. Their nature is unclear at this point. The À' features, assigned to oxygen vacancies in Fig. 14, are mobile in STM images, albeit in a somewhat erratic fashion [116]. Theycan be removed byscanning with a high bias voltage [116,133], hence one needs to consider that theyrepresent adsorbates, speci®callyH atoms adsorbed on a bridging O atom. In [116] this possibilitywas discarded because the densityof these features decreases upon exposure to molecular oxygen, as expected for a `®lling' of vacancies, and because they could not be ¯ashed off. Suzuki et al. [133], however, argued that these features represent H atoms on bridging oxygen atoms because their number density decreases upon irradiation with 20 eV-electrons and increases upon exposure of the surface to atomic hydrogen. A few À' features were resistant against irradiation with electrons and were attributed to H atoms trapped on an oxygen vacancy. These authors concluded that H adatoms, either on bridging oxygen atoms or on O vacancies, still exist on TiO2(1 1 0) when surfaces were prepared bysputtering and annealing. The H was supposed to stem from the bulk. Recent STM of clean and water-covered surfaces showed that oxygen vacancies and hydroxylated bridging oxygen atoms are imaged slightly different in STM images [134], see Fig. 15. The main U. Diebold / Surface Science Reports 48 (2003) 53±229 83

Fig. 15. Ball-and-stick model and corresponding simulated STM image at 1 V showing the appearance of a: (a) vacancy-free surface, (b) bridging oxygen vacancy, (c) bridging OH group and (d) water molecule on top of a ®vefold coordinated Ti atom. Big atoms: O, smaller atoms: Ti, smallest atoms: H. Note that both, a vacancyand an OH group in the bridging oxygenrows, appears as bright spots on dark rows in STM images. From Schaub et al. [134]. # 2001 The American Physical Society. difference in their appearance is their extent in [0 0 1] direction (vacancy6.6 AÊ , OH 4.8 AÊ ), and their apparent height (0.4 AÊ , 0.2 AÊ ) [134]. In this author's opinion, most of `A' features should still be attributable to O vacancies, at least at `fresh' surfaces that have been prepared in a good vacuum. There is overwhelming evidence from spectroscopic measurements that such O vacancies exist, and the increasing experience with STM measurements of slightlydefective surface supports this interpretation. One needs to point out, however, that an (inadvertent) exposure to a few Langmuirs of water (which dissociates at vacancies, hence ®lls them with a hydroxyl) could be suf®cient to replace all O vacancies with two H-covered bridging oxygen atoms. While the calculated difference in image contrast (Fig. 15) gives hope to distinguish between an O vacancyand a H-covered bridging O atom, the stronglytip- and bias-dependent appearance of the `A' features [116,133] in actual images makes their quanti®cation tricky. 84 U. Diebold / Surface Science Reports 48 (2003) 53±229

2.2.1.4.3. Oxygen vacancies created by other means. Oxygen vacancies can also be created by bombardment with electrons. TiO2 is the classic example for a maximum-valencycompound material where electron-stimulated desorption occurs via the Knotek±Feibelman process [135]. Bombardment with energetic electrons creates a core hole in the Ti3p level. With a certain probability, this hole is filled through an inter-atomic Auger process from a neighboring O atom. If two (instead of the usual one) valence electrons are emitted during the Auger decay, the oxygen anion becomes positivelycharged. The previouslyattractive Madelung potential changes into a repulsive one, and an O ion is emitted [136]. This process has a threshold energythat correlates with, but is not exactly located at, the Ti3p edge as discussed in detail in [137]. Such electron-stimulated defects behave somewhat different than thermallycreated ones [138]. It is generallyassumed that electron bombardment results in ejection of bridging oxygen atoms, but direct evidence from STM studies points towards more complicated structures [139]. The high current and high field provided byan STM tip has been used to create protrusions and craterlike depressions structures [140]. Irradiation with high-energyelectrons

(300 keV) induced a TiO as well as an intermediate TiO2-II phase [141]. Defects can also be created byirradiation with UV light, but nothing is known about their structure [142]; as is generallythe case, the cross-section for photon-stimulated desorption is much less as compared to electron-stimulated desorption. Sputtering with rare gas ions reduces the surface oxygen content. Usually, the long-range order of the surface is lost, and the LEED pattern disappears. Spectroscopic measurements as well as adsorption experiments indicate the defects are more complex, involving more than one atom, and are partiallysubsurface [138]. There are indications that sputtering does not completelyrandomize the surface but results in a surface with short-range order that is changed from a twofold to a fourfold symmetry [143]. Generally, sputter-induced damage can be removed easilybyannealing in UHV [74].

2.2.1.4.4. Line defects. STM images of UHV-annealed surfaces (which exhibit a (1 Â 1) LEED pattern) often show dispersed bright stands, typically several tens of AÊ ngstroms long. Theyare distributed across terraces and have a tendencyto grow out of step edges onto the lower terrace (see Fig. 16a). The strands are centered on top of bright rows of the lower terrace (on top of the fivefold coordinated Ti atoms). STM often shows a bright spot at the end. A double-strand structure is resolved in high-resolution images (Fig. 16a). As shown in Fig. 16, these strands are precursors for the (1 Â 2) reconstruction. Conflicting geometric models have been proposed for this reconstruction. These are discussed in Section 2.2.2. The presence of such dispersed strands is sample-dependent. Li et al. heated samples cut from the same specimen to different temperatures in a furnace in order to achieve different levels of bulk reduction (see Fig. 5). After sputtering and annealing at 973 K for 10 min, strands were present on dark blue samples. Less reduced samples that exhibit a lighter color did not show anystrands [144]. The reduction state of the crystal may not only in¯uence the density but also the geometric structure of the strands [76]. Investigations with numerous TiO2 samples in this author's laboratoryhave shown that small amount of bulk impurities (well below the detection limit of commonlyused surface analytical techniques) can also cause strands on the surface. This is in addition to the bright strings caused byCa segregation discussed in the next section.

2.2.1.4.5. Impurities. Commercial TiO2 single crystals are generally quite clean. A common impurity that has been investigated on TiO2(1 1 0) is calcium. It tends to segregate to the surface upon high- temperature annealing [145±148]. Typically, Ca can be depleted from the near-surface region in a few U. Diebold / Surface Science Reports 48 (2003) 53±229 85

Fig. 16. (a) STM image of strings growing out of the upper terrace. These strings are centered at the ®vefold coordinated Ti rows of the lower terrace. Theyare precursors of the (1 Â 2) reconstruction. The surface was annealed at 1020 K for 1 min (8:5nmÂ 12:3 nm, sample bias: 0.7 V, tunneling current: 0.3 nA). (b) Variable current scan of the (1 Â 2)-ordered strings on the surface annealed at 1150 K (70 nm Â 70 nm, sample bias 2.0 V, tunneling current 0.3 nA). The vertical axis in both images is parallel to the [0 0 1] direction. From Onishi and Iwasawa [123]. # 1994 Elsevier. sputter/annealing cycles. For high coverages, it forms a well-ordered overlayer which can clearly be observed in LEED [145±147]. Zhang et al. reported a 60 31 structure in LEED and [0 0 1]-oriented, bright strands in STM. The formation of a CaTiO3-like surface compound was tentativelyproposed. This structure was questioned byNo Èrenberg and Harding [147], who reported a p(3 Â 1) reconstruction and antiphase boundaries which could give rise to the LEED pattern observed in [145]. Model calculations suggest that Ca substitutes ®vefold coordinated surface Ti atoms. Two vacancies form at the sites of the threefold coordinated O atoms located next to the Ca atom. Ordering is energeticallyfavorable, and the relaxed structure is predicted to be buckled [147]. The Ca-segregated surface is very¯at which points to a rather low surface free energy.Such a ¯attening takes place also in other segregated oxide systems [149]. Ca segregation also appears to affect the formation of CSPs (see next section). It has been reported that plasma treatment is ef®cient for the removal of Ca at low temperatures [150]. The presence of H on nominallyclean surfaces has been discussed in the context of STM of O vacancies, above. In addition to Ca, some Mg segregation is sometimes observed with LEIS. Persistent Al impurities were detected with static SIMS [151]. This technique is verysensitive to certain elements. The Al probablyresulted from the polishing procedure. SSIMS measurements have also shown that the most common impurityin samples from different vendors is K [152]. All alkali and earth alkali impurities can be removed to a large extent bysputtering/annealing cycles.A persistent impurity,which was onlyapparent in STM images, was attributed to V contamination [153].

2.2.1.4.6. Crystallographic shear planes. As pointed out above, the oxygen loss through thermal annealing that occurs in the bulk of TiO2 crystals can lead to CSPs. An overview of CSPs and their 86 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 17. (A) STM image (400 Ð Â 400 Ð, 0.2 nA, 1 V, 773 K) showing a pair of CSPs running in the h335i direction across (1 Â 2)-reconstructed terraces. (B) Schematic diagram of a h001i projected view of a pair of CSPs terminating at the [1 1 0] surface. Oxygen ion distorted octahedra centered on the Ti4 ions (dark dots) are indicated bythe diamond shapes (cf. Fig. 6). The interstitial Ti populates the h001i channels in the bulk lattice as indicated at the bottom left while a normal step edge is shown in the top left corner. The distorted octahedra forming the CSPs are shown in grayand have not been relaxed from their bulk positions in the normal and displaced lattices. Where the CSPs terminate at the surface 1/2 height steps are formed which have differing structures. From Bennett [157]. # 2000 Royal Society of Chemistry. relation to surface structure is given byBennett et al. [87]. The CSPs are formed byshifting the normally edge-sharing octahedra (see Fig. 6a) to a face-sharing arrangement. Theycan be thought as consisting of small, undisturbed volumes of the regular rutile lattice, separated byCSPs (see Fig. 17). As their concentration increases, theycan order into a regular arrayknown as Magne Âli phases (see Fig. 4). Two homologous series of MagneÂli phases are known to exist with shear planes along {1 2 1} and {1 3 2} directions (see Table 1 in [87]). The CSPs intersect the (1 1 0) surface with adjacent sections of the crystal being displaced by1.6 AÊ . A good example of CPSs intersecting with a (1 Â 2)-reconstructed surface is shown in Fig. 17.

One of the ®rst STM studies of reduced TiO2 showed various periodic structures which were interpreted as CSPs [84,154], but these images look quite different from later investigations of CPSs. TiO2 crystals that were subjected to heat-treatment only (without sputtering) showed Ca segregation and step edges along 110, 111 and 111 directions [85]. These were interpreted as CSPs belonging to a close packed familyof {1 1 2} planes [85]. The substantial Ca segregation mayhave in¯uenced this structure.

The surface structure of the most oxygen de®cient MagneÂli phase, Ti4O7, was investigated also [155]. After repeatedlysputtering and heating (1223 K) of a TiO 2(1 1 0) crystal (which diminished the initially observed Ca segregation), Bennett et al. [87] reported STM and LEED results of CSPs on a crystal with dark blue/black color. The crystal showed a rippled texture that was visible with the naked eye. U. Diebold / Surface Science Reports 48 (2003) 53±229 87

Fig. 18. Models for the TiO2(1 1 0)-(1 Â 2) surface. Small white balls: Ti, large black balls: O. (a) The `missing row' model, obtained byremoval of one row of bridging oxygens,was originallyproposed byMùller and Wu [111]. This model is inconsistent with more recent STM images and ®rst-principles calculations. (b) The `added-row model' has Ti2O3 stoichiometry. It was suggested by Onishi et al. [122]. (c) The `missing unit' model was proposed byPang et al. [163]. Recent evidence suggests that the structures in (b) and (c) might be both present at TiO2(1 1 0)-(1 Â 2) for different conditions. From Tanner et al. [113]. # 1998 Elsevier.

The Ti (390 eV)/O (510 eV) AES ratio was 1.22 for the heavilyreduced surface 2.Thisistobecompared to a ratio of 1.14 for the (1 Â 1)-terminated surface. On different parts of the crystal, streaks along the 1 11 and 1 12 directions, or a superposition of the two, were observed with LEED. STM images show a large concentration of steps running along h112i and h1 12i with the expected step height of 1.6 AÊ ,as well as a strong variation in mesoscopic morphology [87]. The CSPs also act as nucleation sites for re- growth of new TiO2 layers during high-temperature oxidation (see Section 2.2.2) [157].

2 Probablythe color of the sample in [87] was verydark. No CSPs have observed on anyof the cubes shown in Fig. 5 [156]. 88 U. Diebold / Surface Science Reports 48 (2003) 53±229

2.2.2. Reconstructions

2.2.2.1. Reconstruction under reducing conditions: the structure(s) of the (1 Â 2) phase. The most commonlyobserved reconstruction on TiO 2(1 1 0) surfaces has a (1 Â 2) symmetry with a doubling of the periodicityalong the 1 10 direction. Various models have been suggested for this reconstruction and the most popular ones are depicted in Fig. 18 [113]. A(1Â 2) LEED pattern was originallyobserved after high-temperature annealing of a reduced

TiO2(1 1 0) sample in ultrahigh vacuum (UHV). Based on Ti:O AES ratios it was interpreted as alternate rows of bridging oxygen missing from the regular (1 Â 1) surface (``missing-row model'' [111], Fig. 18a). One of the ®rst atomicallyresolved STM results on this surface was also interpreted as missing bridging oxygen rows [114,158]; however, the Ti atoms underneath the missing oxygen atoms had to be shifted byhalf a unit cell in [0 0 1] direction to account for the observed image contrast. A structure with a (3 Â 2) symmetry was reported in [159]. A model for this reconstruction was discussed where this symmetry is achieved by removing 1/3 or 2/3 of the oxygen in the bridging oxygen rows. However, such a reconstruction has not been reported byother groups. The simple missing row model for the (1 Â 2) structure in Fig. 18a has been abandoned on the basis of more recent results. In STM the (1 Â 2) reconstruction is commonlyobserved as a series of bright strings along the [0 0 1] direction [113,114,123,158,160±162], see Fig. 16. At low coverage, the strings grow preferentiallyout of the upper terrace onto the lower one ( Fig. 16a) [123]. At ®rst, theyare scattered across the terraces with a minimum distance of 13 AÊ along the [0 0 1] direction. Theyconsist of bright double strings (although the double-ridge structure is often not resolved), with a bright dot at the end. Antiphase boundaries are observed in high-resolution images of a fullydeveloped (1 Â 2)- reconstructed surface [113]. Higher periodicities, i.e., a local (1 Â 3) reconstruction, have also been observed [161,163,164]. In STM images the (1 Â 2) strands generallyhave an apparent height smaller Ê than a regular TiO2 step edge of 3.2 A, and are in registrywith the bright rows of the (1 Â 1) substrate. Because most researchers report empty-state images and because these are dominated by the tunneling into mostlyTi3d-derived states (see Section 2.2.1.3), bright strands in line with bright substrate rows implythat the (1 Â 2) strands are at the position of ®vefold coordinated Ti atoms and not at the bridging oxygen atoms. STM images of a simple missing row structure are expected to show a bright feature above the missing bridging oxygen row (provided the STM tip is a reasonable distance from the surface [112]), inconsistent with the registryobserved experimentally.The rows can be removed bytunneling under èxtreme conditions' (Vs 1:5V,I 10 nA [113]). First-principles total-energycalculations show that the added Ti2O3 model (discussed next) is energeticallyfavored [165] and that the missing row structure is energeticallyequivalent to a (2 Â 1) structure (where everyother bridging oxygenis removed) [99]. For all these reasons, the missing-row reconstruction is no longer considered a viable model. Earlyon, Onishi et al. [122,123] suggested a quite different model. It consists of double rows of Ti cations in a distorted tetrahedral con®guration (Fig. 18b). The structure has Ti2O3 stoichiometry, and the model is often called àdded Ti2O3 rows'. However, it needs to be emphasized that the structure does not resemble the one found in the corundum Ti2O3 structure. Rather, the Ti cations reside in positions similar to interstitial sites in the rutile lattice [91]. Self-consistent total-energyand electronic structure calculation found that this added `Ti2O3' row structure has a lower surface free energythan the missing row structure and that it is consistent with the contrast in STM [165]. Recent VASP calculations show that such strands can be added at low energycost [91], but also that manyother U. Diebold / Surface Science Reports 48 (2003) 53±229 89 con®gurations are energeticallylikely.The model is also supported byESDIAD [166], high-resolution STM [113], and ion scattering [164] measurements. Based on the fact that (1 Â 2) rows extend out of step edges, a modi®ed model of the missing row structure has been proposed byMurrayet al. [88] which involves narrow rows with missing bridging oxygens that are effectively part of the upper terrace. Lateral relaxations were also included [88]. This model was shown to be consistent with calculated surface charge densities [112]. Based on the same scheme, an àdded-row model' was proposed byPang et al. [163] for the fullyreconstructed surface.

This consists of narrow, long regions of the regular TiO2 structure with all the atoms in bulk-like positions and with all bridging oxygen atoms missing (Fig. 18c). The black grooves between the bright rows are then due to the missing TiO2 units separating the rows. (Consequently, this model has also been called `missing unit' structure [113].) The stoichiometryof the added rows is Ti 3O5. The model was based on STM images with unusuallyhigh resolution and the observation of a (1 Â 3) phase consisting of thicker rows. Charge-densitycalculations byPang et al. supported the added-row model, but were inconsistent with either the added Ti2O3 or the simple missing row model (Fig. 18a and b, respectively). The off-normal lobes in ESDIAD images were supposed to stem from the O atoms at either side of the added `rows', adjacent to the missing units. Pang's model has been questioned byTanner et al. [113,167,168]. The (1 Â 2) strands that are part of `restructured' surfaces after low-temperature oxidation (see Section 2.2.2.2) are consistent with the added Ti2O3 model rather than the added Ti3O5 row model [144]. Ion scattering measurements are also consistent with the added `Ti2O3' model [164] although (somewhat surprisingly) extra oxygen atoms at the position of the ®vefold coordinated Ti atoms were postulated according to these measurements. In a recent paper, several energeticallyaccessible reconstructions were considered [91]. As is discussed in the next section (2.2.2.2), the two added-row models appear not to be mutuallyexclusive, and the formation of one or the other structure mayjust depend on the sample preparation parameters and crystal reduction state of the crystal.

2.2.2.2. Restructuring under oxidizing conditions. The first atomic-level investigation of the dynamic processes that occur when reduced TiO2 crystals are exposed to oxygen was reported by Onishi and Iwasawa [169]. These authors used a blue crystal with a resistivity of 2 O m, i.e., with a color probably comparable to cube 5 in Fig. 5b, see Table 2. Theyacquired STM images while the sample was kept at a À5 temperature of 800 K and under an O2 background pressure of 1 Â 10 Pa. Added rows (interpreted as Ti2O3 rows, Fig. 18b) and `hill-like features', appeared while imaging the surface, and disappeared when the oxygen was pumped out from the chamber. This effect was not tip-induced; the same structures were observed on areas of the sample that were not imaged during the high-temperature oxygen exposure. It is now established [75,76,156,170±172] that such an oxygen-induced surface restructuring effect is attributed to the reoxidation of the reduced crystal, as already suggested by Onishi and Iwasawa [169]. As mentioned above, the bulk of sub-stoichiometric TiO2Àx samples contains, in addition to O vacancies, Ti interstitials which show a high diffusitivityat elevated temperatures. When these interstitials appear at the surface, theycan react with gaseous oxygenand form additional TiO 2 (or TiaOb) structures. For extreme cases a complete reoxidation of the whole crystal can be achieved, see cube 2 in Fig. 5 that has been re-oxidized to a transparent color. This reoxidation process has pronounced effects on the surface structure. The kinetics of the oxygen-induced restructuring process as well as the resulting surface morphologies depend on sample temperature, annealing time, gas pressure, and reduction state (i.e., 90 U. Diebold / Surface Science Reports 48 (2003) 53±229

18 À6 Fig. 19. STM images 500 Ð Â 500 Ð of a TiO2 (1 1 0) surface taken at room temperature. O2 (1 Â 10 mbar) was dosed at (a) 500 K, (b) 520 K, (c) 550 K, (d) 660 K for 10 min, (e) 710 K for 15 min, and (f) 830 K for 20 min. Before each gas exposure, the sample was sputtered and annealed in UHV at 880 K for 30 min which renders ¯at, (1 Â 1)-terminated surfaces. From Li et al. [75]. # 1999 Elsevier.

`age' or color) of the crystal. These parameters have been investigated in detail by Li et al. À6 [75,144,156,170,171]. For example, Fig. 19 shows the effect of annealing in 1 Â 10 mbar O2 at various temperatures [75]. Before each gas exposure, a ¯at (1 Â 1)-terminated surface was prepared by sputtering and annealing in UHV. The surface morphologyof the oxygen-exposed sample is very temperature dependent. For medium temperatures, surfaces are relativelyrough with manysmall-scale 18 features. Isotopicallylabeled O2 gas was used for the annealing excursions. In SSIMS and low-energy He ion scattering measurements, the signal from 18O atoms can clearlybe separated from the (naturallymuch more common) 16O isotope in the crystal. All the surfaces in Fig. 19 showed an enrichment with 18O, with a maximum of 18O surface concentration around 660 K. The structures that form for intermediate annealing temperatures (520±660 K) are better seen in the small-scale image in Fig. 20. Theyconsist of small, (1 Â 1)-terminated islands and irregular networks of connected `rosettes', i.e., six bright spots in a pseudohexagonal arrangement, as well as small strands. A model for the rosette network is shown in Fig. 21. It consists of atoms in bulk-like positions with some atoms missing from the regular (1 Â 1) structure. LDA-based ®rst-principles calculations [75] showed that such a rosette structure is stable. The same calculations also predict sizable relaxations. U. Diebold / Surface Science Reports 48 (2003) 53±229 91

Fig. 20. An atomicallyresolved STM image 150 Ð Â 150 Ð of a surface prepared as in Fig. 19. Small (1 Â 1)-terminated islands and patches of connected pseudohexagonal rosettes are seen. From Li et al. [75]. # 1999 Elsevier.

Fig. 21. Atomic model (top and side view) for the oxygen-induced structure observed in (a). A bulk-terminated (1 Â 1) island is shown on the right side and the unit cell is indicated. Small white balls are Ti atoms. Shadowed large balls represent oxygen atoms; darker shading indicates higher z-positions. The rectangle indicates the unit cell of the 1 Â 1 structure. The network patch (`R') on the left side consists of an incomplete TiO2(1 1 0)-(1 Â 1) layer and contains only atoms at bulk position. The strands probablyhave a structure similar to the added Ti 2O3 model in Fig. 18b. From Li et al. [75]. # 1999 Elsevier. 92 U. Diebold / Surface Science Reports 48 (2003) 53±229

The rosette structure can be explained simplybythe formation of (partiallyincomplete) TiO 2 layers through a growth process where the Ti atoms come from the reduced bulk and the 18O from the gas phase. The kinetics of the growth determines the relative concentration of the incomplete structures (the rosettes and strands) and the (1 Â 1) islands on the surface. The surface is ¯at and (1 Â 1)-terminated when the growth is slow in comparison with surface diffusion processes. This can be achieved in various ways, either during annealing at high temperatures, Fig. 19f, or when the ¯ux of one of the constituents is small (in verylight samples with a small concentration of interstitials [156]) or at lower

O2 background pressures. Conversely, on very dark crystals, or at intermediate temperatures, a substantial part of the surface can be covered with rosette networks. Note that 60% of the Ti atoms in the rosettes are fourfold coordinated, whereas the Ti atoms exposed on the (1 Â 1) surface are ®vefold coordinated. Clearlythe possibilitythat such structures can form needs to be taken into account when preparing rutile (1 1 0) surfaces for surface chemistryexperiments (see Section 5). À6 18 The surface structure in Fig. 19e, obtained after annealing in 1 Â 10 mbar O2 at 710 K, shows the presence of (1 Â 2) strands on otherwise ¯at, (1 Â 1)-terminated surfaces. This is in agreement with Onishi and Iwasawa's [169] results described above. From atomicallyresolved images of strands connected to rosettes as well as UHV annealing experiments of restructured surfaces it was concluded

[144] that these strands also have the Ti2O3 structure depicted in Fig. 18b. The dependence of the restructuring (as well as the type of (1 Â 2) reconstruction) on the reduction state of the bulk was resolved byBennett et al. [76]. High-temperature STM studies were performed on two different TiO2 samples. On a dark blue/black crystal (that showed already evidence for CSP formation, i.e., darker than cube 3 in Fig. 5) two different structures were observed (see Fig. 22). The dark and bright strings in Fig. 22 were attributed to added Ti2O3 rows (Fig. 18b) and added rows of a bulk-terminated TiO2 layer (Fig. 18c, but with bridging oxygens at the center of the strands), respectively. The latter structure also appears cross-linked with partial `rosettes' (Fig. 21). Both the added-row structure and the rosettes are just incomplete TiO2 structures that form during the growth of additional TiO2(1 1 0)-(1 Â 1) layers. The authors have published impressive web-based STM `movies' [173] (which can be viewed at http://www.njp.org/) of the growth process that show the cyclic completion of terraces and new formation of the cross-linked added-row structures. In contrast, the dark rows (the Ti2O3 added rows) appeared relativelyunreactive for additional growth.

2.2.3. Recommendations for surface preparation

Although the TiO2(1 1 0)-(1 Â 1) surface is considered the `best-characterized', prototypical metal oxide surface, the above summaryclearlyshows that its atomic-level structure is quite complex. The recent STM results summarized above clearlyindicate that both the oxidation conditions and the history of the TiO2(1 1 0) sample have signi®cant bearing on the morphologyof the surface, the presence of strands, rosettes, or CSPs. The variations in the surface structure with O2 pressure, crystal temperature and bulk defect densityare so vast that one could suspect chemistryof the TiO 2(1 1 0) surface to be signi®cantlyvariant for samples oxidized under different conditions. For example, the issue of whether water is molecularlyor dissociativelyadsorbed on TiO 2(110)[127,128,138,174,175] (see Section 5.1.2) maybe signi®cantlyclouded in the literature because of studies in which the morphologyof the surface was unknowinglydisordered bythe presence of the rosettes and/or strands observed recentlybySTM.

This level of ambiguitymayalso permeate manyother adsorption studies on TiO 2(110). Guidelines of surface preparation of TiO2(1 1 0) can be extracted from recent work [75,76,150,156,172]. If solely(1 Â 1)-terminated surfaces are desired, light blue crystals (as depicted in Fig. 5) should be used. U. Diebold / Surface Science Reports 48 (2003) 53±229 93

Fig. 22. High-temperature STM image of oxygen-induced features on a rather dark, non-stoichiometric TiO2(1 1 0) crystal. The crystal was exposed to oxygen (5:5 Â 10À7 mbar, 833 K), stopped mid-reaction byremoval of the oxygenoverpressure, and imaged at the same temperature. The two types of strands with different apparent height (dark strands (DS) and bright strands (BS)) are attributed to the presence of different (1 Â 2) structures. The added Ti2O3 rows account for the dark strings. The bright strands are interpreted as a slightlymodi®ed (addition of bridging oxygensat the strands) added-row structure (Fig. 18c). Additionally, bright rows and vacancies are visible on the 1 Â 1 surface (marked BR and V, respectively). Line pro®les are taken in the fast scan direction to minimize thermal drift and tip change problems. From Bennett et al. [76]. # 1999 The American Physical Society.

Annealing in oxygen will then result in stoichiometric (1 Â 1)-terminated surfaces. On the other hand, more complex morphologies with a range of different coordination sites can be formed if dark crystals are annealed in oxygen. Oxygen vacancies, achieved after UHV annealing, can possibly be quenched by exposing to O2 gas. However, one has to be careful that the O2 gas is free of water which adsorbs quite readily(see Section 5.1.2)andthatO2 dissociated at vacancies can result in O radicals (see Section 5.1.3). The rich arrayof surface structures achievable on TiO 2(1 1 0) mayprovide a playground for surface science experiments where the in¯uence of different adsorption sites can be tested.

2.3. The structure of the rutile (1 0 0) surface

2.3.1. The TiO2(1 0 0)-(1 Â 1) surface The rutile (1 0 0) surface has received considerablyless attention than the (1 1 0) crystalface. The rules of autocompensation and creation of non-polar surfaces discussed above (Section 2.2.1.1) allows a 94 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 23. (A) Geometryof the unreconstructed TiO 2(1 0 0)-(1 Â 1) surface. This surface results when the same number of Ti ! OasO! Ti bonds are broken in a bulk crystal, see dashed line in (B). The formation of the 1 Â 3 microfaceted surface, originallyproposed byZschak et al. [186] is displayed in (B). Removal of the volume labeled g produces a stoichiometric surface with {1 1 0} facets. In order to reconcile the reduced character of the 1 Â 3 surface (observed in photoemission), the outermost bridging oxygen atoms (g0) are thought to be missing as well [187]. The transition from the (1 Â 1) surface to the 1 Â 3 microfaceted surface was proposed to occur via breaking the bonds a which leads to the indicated relaxation and creation of an intermediate 1 Â 3 phase [176]. (C) A surface X-raystructure analysisbyZajonz et al. [192] implied a modi®ed, heavilyrelaxed model. (D) The glancing angle X-raydiffraction data that had lead to the original microfacet model was re-evaluated byLandree et al. [193] and a new structure has been proposed. The octahedra model schematic representations in (D) show the microfacet model (top) and the new model (bottom) (from Landree et al. [193]). straightforward prediction of the stable surface termination (Fig. 23A). Again, the same number of Ti ! OasO! Ti need to be broken, as indicated bythe line in Fig. 23B. This results in a strongly corrugated surface, with rows of bridging oxygen atoms at the outermost, (1 0 0)-oriented ridges (Fig. 23A). Indeed a (1 Â 1)-terminated LEED pattern is observed on this surface after sputtering and annealing, and STM and non-contact AFM images are consistent with this model [176,177]. Several theoretical calculations have determined likelyrelaxations of the (1 Â 1) surface [68,101,178,179].In[178] different theoretical approaches and basis sets were tested. All these calculations agree in the general motions of the atoms, although the amount of relaxations differ somewhat. As expected from symmetry, no relaxations occur along [0 0 1]. In the [1 0 0] direction only the ®vefold coordinated Ti atoms show appreciable (downwards) relaxations [68]. Substantial relaxations occur along the [0 1 0] direction with the twofold coordinated and the threefold coordinated oxygen atoms moving in opposite direction of the ®vefold and sixfold coordinated Ti atoms. In Fig. 23A, O atoms would move to the right and Ti atoms to the left. The net effect of these displacements is to increase the effective coordination of the ®vefold coordinated Ti atoms [178]. U. Diebold / Surface Science Reports 48 (2003) 53±229 95

No experimental data on relaxations of the TiO2(1 0 0)-(1 Â 1) surface exist. X-rayphotoelectron and Auger electron diffraction were performed but are insensitive to the details of the surface structure [180].

2.3.2. Reconstructions

2.3.2.1. The microfacet model of the rutile TiO2(1 0 0)-(1 Â 3) surface. In addition to the (1 Â 1)- terminated surface, a (1 Â 3)-reconstructed surface forms relativelyeasilyupon annealing to high temperatures in UHV. It is clearlyevident from photoemission experiments that this surface is partially reduced [27,181±183]. Initially, the observed reconstruction was interpreted as removal of every third row of the outermost, `bridging' oxygen atoms [183]. Such a proposal was verymuch in line with the initiallyproposed [184] (and now largelyabandoned) `missing row' model for the (1 Â 2) structure of the

TiO2(1 1 0) surface (see Fig. 18a). The step structure on this surface maylead to a misinterpretation as additional (1 Â 5) and (1 Â 7) reconstructions as pointed out byMuryn [183]. The first STM work on a reconstructed TiO2(1 0 0) surface was reported byClark and Kesmodel [185]. A glancing angle X-ray diffraction and low energyelectron diffraction study [186] suggested a `microfaceted' model, as shown in Fig. 23B. Removal of the volume assigned with g in Fig. 23B creates facets of the lowest energy(1 1 0) planes. Note that, again, the same number of O ! Ti as Ti ! O bonds are broken. This results in a stoichiometric surface, with twofold coordinated bridging oxygen atoms at the outermost ridges, as is the case for the TiO2(1 1 0) surface. The unit cell in [0 1 0] direction is three times wider. Atomicallyresolved STM images showed bright ridges with a geometryconsistent with the microfaceted model [187,188]. The apparent height between top and bottom of the reconstruction was measured as 3 AÊ with STM, instead of the expected 5 AÊ . This has been associated with tip effects [187,188]. Scanning tunneling spectroscopy(STS) measurements showed considerable difference between dI/dV spectra taken on and in between the bright ridges, respectively [187,189]. This was interpreted as a consequence of missing oxygen atoms on the outermost ridges (i.e., removal of the oxygen atoms labeled g0 in Fig. 23C). This results in a surface termination with threefold coordinated Ti atoms in the outermost plane, consistent with the observation of a reduced surface in photoemission [183]. A photoelectron diffraction studyof the Ti3p level was performed byHardman et al. [182]. The PED curves of the (curve ®tted) Ti3 feature were evaluated. These Ti3 features are supposed to come from Ti atoms located next to oxygen vacancies, and diffraction effects should give information about their surface geometry. Three different positions for vacancies on the microfaceted (1 Â 3) surface were tested. The most likelycon®guration has the missing oxygenatoms at the outermost ridge of the (1 1 0) facets [182], consistent with the STS measurements. In this context, it is also interesting to note that the reduction state of the substrate plays a role in the formation of the (1 Â 3) reconstruction. Almost stoichiometric, Nb-doped TiO2(1 0 0) ®lms are thermallymuch more stable than reduced TiO 2 substrates and do not reconstruct at temperatures where reduced TiO2 substrates alreadyshow a clear (1 Â 3) structure [190]. A model involving `discrete bond breaking' was proposed for the formation of the (1 Â 3) surface [176,177]. Surfaces were prepared that exhibited both, the (1 Â 1) termination and the ridges typical for the (1 Â 3) surface. STM and non-contact AFM images showed an intermediate phase, which had (1 Â 3) symmetry, but did not possess the characteristics of the microfaceted structure. Raza et al. [176] suggested that the bonds labeled a in Fig. 23B are broken, which allows the Ti atoms to relax towards the other rows of bridging oxygen atoms. 96 U. Diebold / Surface Science Reports 48 (2003) 53±229

2.3.2.2. Is the simple microfacet model valid?. Despite the experimental results discussed in the previous section, it is presentlynot clear if the microfacet model for the (1 Â 3) surface (Fig. 23B) is valid. In fact, recent evidence indicates that it mayrepresent an oversimplification. Theoretical calculations are not in agreement with the microfacet model. A tight-binding calculation derives a higher surface energyfor the relaxed, microfaceted surface as compared to the (1 Â 1)- terminated one [191]. DFT ab initio calculation [179] also showed a considerablyhigher surface free energy. This implies that there would be no driving force for the reconstruction, at least not at 0 K and under UHV conditions. These surprising results were attributed to the fact that stoichiometric surfaces were considered in the calculations (i.e., the oxygen atoms g0, Fig. 23C, were not removed) while experimental spectroscopic data show that the (1 Â 3) surface is partiallyreduced. As was pointed out byLindan et al. [179], the Ti3 state associated with the reduction should correctlybe treated with spin- polarized DFT calculations. A recent grazing incidence X-raydiffraction (GIXD) analysisshows strong lateral and vertical relaxations (1 AÊ and more) of the titanium and oxygen atoms in the top layer [192]. The coordination of the surface Ti atoms differ considerablyfrom the simple microfacet model, see Fig. 23C. Threefold coordinated Ti atoms were found at the facet ridges (Ti1 in Fig. 23C), in agreement with previous work. In addition, various Ti atoms were found in different con®gurations, e.g. Ti3 in an oxygen bridge site, titanium B in a trigonal prismatic coordination, as well as an interstitial Ti site A. Obviously, this model deviates stronglyfrom the simple microfaceted structure. The original GIXD data, which were the basis for the microfaceted model, were re-evaluated recently byLandree et al. [193].In[186] the data were interpreted using Patterson functions which show only interatomic vectors, not the `true' atomic positions. The re-evaluation was based on a technique known as `direct method', in essence a Fourier transform of the measured data in connection with a search for possible (unknown) phases [194]. It was found that the microfacet model gave verypoor agreement with the data as compared to a model that contained four Ti and six to eight O atoms in the surface unit cell. In this model, the Ti atoms reside in edge- and corner-sharing octahedral units, as opposed to the normal rutile structure which is composed of corner-sharing octahedra only(see Fig. 23D). The reconstructed structure is rationalized as a standard con®guration for non-stoichiometric defects such as CSPs, see Fig. 17. The reduced states observed with spectroscopic measurements would then be accommodated similar as in the bulk.

The surface unit cell of the TiO2(1 0 0)-(1 Â 3) structure is quite large with manyatoms, hence it is no surprise that it is dif®cult to determine conclusivelythe exact surface geometry.While the electron diffraction [182], STM and AFM results [176,177,187±189], together with ESDIAD measurements [195] are consistent with the microfacet model of the (1 Â 3) surface, none of these techniques gives direct evidence of surface geometries. The (1 Â 1) and (1 Â 3) reconstructions provide a convenient system for site-sensitive surface chemistry experiments, because one can reversibly cycle a TiO2(1 0 0) crystal between the (1 Â 1) surface and the reconstructed one [196,197]. Hence, it would be quite important to have additional experimental as well as theoretical support for one of the models currently suggested and depicted in Fig. 23.

2.4. Rutile (0 0 1)

There is onlyone wayto cut a rutile crystalin (0 0 1) direction ( Fig. 24). Although this creates a non- polar, autocompensated surface, it does not represent a low-energycon®guration. It becomes clear U. Diebold / Surface Science Reports 48 (2003) 53±229 97

Fig. 24. Surface termination of the rutile TiO2(0 0 1) surface. Onlyone possibilityexists to cut a TiO 2 crystal in this direction, see the side view at the left side. Surface Ti atoms are fourfold coordinated and surface O atoms twofold coordinated. immediatelywhen reviewing the coordination of the surface atoms. All the Ti atoms are fourfold coordinated, and all the O atoms twofold coordinated. Hence the number of broken bonds on this surface is higher than on the other low-index rutile surfaces discussed so far. Consequently, the (0 0 1) surface has a high surface energyand tends to facet or reconstruct. Based on LEED studies, {0 1 1} and {1 1 4} facets have been identi®ed byTait and Kasowski [198] and Firment [199]. Poirier et al. [200] annealed a crystal at 400 8C and performed STM and LEED studies. No atomic resolution was achieved, but several other crystal planes were additionally identi®ed. AFM images showed relatively ¯at surfaces [201], but the resolution in the images was not high enough to identifythe atomic structure of the facets. A recent STM studywith near-atomic resolution byFukui et al. [202] showed the different structures that evolve when a sputtered surface is heated to increasinglyhigher temperatures, see Fig. 25. Similar preparation conditions are usuallyused to prepare surfaces with different `faceted' terminations. The preparation conditions for Fig. 25b, which should give a (1 Â 1) surface, showed hills instead. Some parts of these an average {0 1 1} orientation, but no ordered LEED was observed [202]. (In this context it might be worth mentioning that a rutile crystal, cut to expose a (1 0 1) surface, showed faceting upon heating [152]. Thus the formation of facets which actuallyexpose an unreconstructed (1 0 1) surface is not likely.) Heating to a temperature of 1050 K (Fig. 25c) showed well-resolved rows, running along the 110 and 1 10 directions. High-resolution STM showed that these rows consist of a staircase of narrow terrace (a `bleacher-like structure'). The (unreconstructed) {1 1 4}-faceted model of the

TiO2(1 0 0) surface consists of a narrow terrace of the bulk-terminated (0 0 1) surface with steps of (1 1 1) orientation [202]. STM images of the rows are not consistent with this structure, although the average slope was {1 1 4}. The image in Fig. 26 shows the drastic changes that can occur upon veryhigh-temperature annealing of TiO2(0 0 1). The micrograph has been taken with an optical microscope after heating a TiO2(0 0 1) crystal for 1 h at 1300 8C in UHV [203]. The lines are caused byslip. A non-equilibrium structure was observed after rapidlyquenching a TiO 2(0 0 1) crystal from a similar high temperature [204]. The most detailed structural results on TiO2(0 0 1) have been described below. The rutile (0 0 1) surface is the one where the least detailed structural information is available. This is particularlyunfortunate, as the `faceted' surface has been used extensivelyfor the studyof reactions of 98 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 25. STM images (100 nm Â 100 nm) of TiO2(1 0 0) surfaces depending on annealing temperature after Ar -ion sputtering. (a) Before annealing. Variable current image (Vs 2:0V,It 0:2 nA). (b) After annealing at 970 K for 5 min. Some parts of the hill-like structures have an average plane of {0 1 1}, but most of the surface is disordered and no well- ordered pattern was observed byLEED. (c) After annealing to 1050 K for 5 min. A well-contrasted LEED pattern was observed that was previouslyassigned to a {1 1 4} structure [199]. The surface is not microfaceted but instead consists of rows which show a `bleacher-like' structure in high-resolution images. The average slope is identical to a {1 1 4} face. (d) After annealing at 1160 K for 5 min the surface is covered with 5 nm high particles. The STM images in (a, b, d) are variable current images, the one shown in (c) a topographic image. From Fukui et al. [202]. # 2001 Japanese Societyof Applied Physics.

Fig. 26. A TiO2(0 0 1) surface after annealing at a veryhigh temperature (1300 8C, 1 h in UHV). The image was taken with an optical microscope. It shows lines due to slip along certain crystallographic directions. From NoÈrenberg et al. [203]. # 2000 Elsevier. U. Diebold / Surface Science Reports 48 (2003) 53±229 99 organic molecules, see Section 5.2. The (0 0 1) surface is also the crystal orientation of choice for electrochemical studies, as the electrical conductivityis highest along the [0 0 1] direction. The recent work discussed here clearlyshows that there are inconsistencies with the established interpretation of reconstructed TiO2(0 0 1). Additional theoretical and experimental work that could help to resolve the geometryof this surface would be quite valuable.

2.5. Vicinal and other rutile surfaces

Vicinal surfaces of TiO2 have not been studied extensively. A study of Na adsorption on a stepped (4 4 1) surface was reported byOnishi et al. [205]. Unpublished experiments from this author's laboratorywith a similarlycut crystalshowed macroscopic faceting upon annealing.

The most detailed investigation was performed recentlyon a TiO 2(210)surface [206]. In a formal sense, TiO2(2 1 0) lies midways between (1 1 0) and (1 0 0), and is the most simple vicinal surface. Atomistic simulations, based on Coulombic interaction between ions and a short-range repulsive interaction, predicted an asymmetric sawtooth-like structure of the surface, consisting of {1 1 0} nanofacets. The widthp of each nanofacet is 1.5 times the width of the surface unit cell of the (1 1 0)-(1 Â 1) structure (i.e. 3a= 2). The nanofacets terminate with a row of Ti atoms carrying bridging oxygen atoms. The surface energyof this structure is predicted to be 2.07 J/m 2. (This is to be compared to a surface 2 energyof 1.78 J/m derived using a similar calculation for TiO2(110) [206].) STM images showed a (1 Â 1)-terminated surface that could be consistent with this structure, although the interpretation was again made dif®cult bybalancing electronic effects with the verystrong corrugations of this surface.

The structure of the TiO2(1 1 1) surface was investigated byOnishi and co-workers [207] with LEED and STM. Depending on the annealing temperature, the surface shows a varietyof reconstructions.

2.6. Anatase surfaces

Most commercial titania powder catalysts are a mixture of rutile and anatase (e.g. the most often used Degussa P25 contains approx. 80±90% anatase and the rest rutile [11]). For certain photocatalytic reactions and non-photoinduced catalysis such mixtures work best [208]. There is growing evidence that anatase is more active than rutile for O2 photo-oxidation, but not necessarilyfor all photocatalytic processes. Anatase behaves differentlythan rutile in gas-sensing devices, and most photovoltaic cells are based on granular thin ®lms with an anatase structure [29]. Anatase and rutile show inherent particle size differences and this might cause some of the observed differences in chemical properties. However, in order to gain a better understanding of TiO2-based devices, it is clearlyimportant to obtain atomic- scale information on well-characterized anatase surfaces. High-purity titania powder catalysts are typically made in a ¯ame process from titanium tetrachloride [11]. Manyadditional synthetictechniques processes have been applied [209±213]. The shapes of the crystallites vary with preparation techniques and procedures. Typically, (1 0 1) and (1 0 0)/(0 1 0) surface planes are found, together with some (0 0 1) [70]. Several theoretical studies have predicted the stabilityof the different low-index anatase surfaces [70,214,215]. The (1 0 1) face is the thermodynamically most stable surface, see the calculated surface energies in Table 4, [216,217]. While it is dif®cult to obtain accuracyfor surface energies numbers with DFT calculations, the relative surface energies in Table 4 should still be meaningful. The calculated Wulff shape of an anatase crystal, based on these values, compares well with the shape of naturallygrown mineral samples, see Fig. 27. 100 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 4 2 a Comparison of calculated surface formation energies (J/m ) for relaxed, unreconstructed TiO2 surfaces Rutile (1 1 0) Anatase (1 0 1) (1 0 0) (0 0 1) (1 0 3)f (1 0 3)s (1 1 0) 0.31 0.44 0.53 0.90 0.84 0.93 1.09 a Two different structures for the (1 0 3) surfaces (a `faceted' and a `smooth' one) have been considered. From [216,217].

Interestingly, the average surface energy of an equilibrium-shape anatase crystal is smaller than the one of rutile [216,217], which might explain the fact that nanoscopic TiO2 particles are less stable in the rutile phase. Experimental investigations on single-crystalline anatase are just starting. Meaningful surface science investigations necessitate single-crystalline samples. While rutile crystals are readily available, suf®cientlylarge and pure anatase crystalsare more dif®cult to obtain. Because anatase is a metastable phase, it transforms into rutile at relativelylow temperatures [210], with the transition temperature dependent on impurities, crystal size, sample history, etc. From the recent progress in synthesizing single-crystalline anatase samples with high purity [209] as well as the successful growth of epitaxial thin-®lms on appropriate substrates [218±221] one can expect a rapid increase in the interest in anatase

TiO2 in the near future.

2.6.1. Anatase (1 0 1) Two reports on the structure of anatase (1 0 1) surfaces have appeared veryrecently [220,222]. The (1 0 1) surface on an anatase sample grown bychemical transport [209] showed a (1 Â 1) surface after mild sputtering and annealing [222]. A mineral sample similar to the one displayed in Fig. 27 was used in an STM studybyHebenstreit et al. [220]. In order to avoid the contaminations in this natural single crystal, a 700 AÊ thick, epitaxial ®lm was grown on the surface, as described in [221]. Sputtering and

Fig. 27. (a) The equilibrium shape of a TiO2 crystal in the anatase phase, according to the Wulff construction and surface energies calculated in [216] (from Lazzeri et al. [216], # 2001 The American Physical Society). (b) Picture of an anatase mineral crystal. U. Diebold / Surface Science Reports 48 (2003) 53±229 101

Fig. 28. (A) STM results of an anatase (1 0 1) single crystal. The monoatomic terraces terminate with step edges that run predominantlyin certain preferred orientations. (B) Atomic models (side and top view) of the anatase (1 0 1) surface. From Hebenstreit et al. [220]. # 2000 The American Physical Society. annealing again produced a (1 Â 1) termination in LEED. The surface has onlya pm symmetry,giving rise to a preferential orientation of step edges (Fig. 28A). Based on the rules of autocompensation for step edges (see Section 2.2.1.1) a reasonable model for steps is given in Fig. 28 [220]. Titanium atoms at the terraces have ®vefold and sixfold coordination, and titanium atoms at the step edges are fourfold coordinated. These have a higher reactivityagainst gas adsorption [220]. Twofold coordinated oxygen atoms are located at the ridges of the saw tooth-like structure. According to [70], theyrelax inwards by 0.21 AÊ . The threefold coordinated O atoms relax outwards by0.06 AÊ and the ®vefold coordinated Ti atoms inwards by 0.17 AÊ , so that the surface exhibits a slightlybuckled geometry.The tunneling site in the atomic-resolution image in Fig. 29 probablyextends across both, the twofold coordinated oxygen atoms and the ®vefold coordinated Ti atoms. In an image taken with a higher tunneling current (12 nA), where the tip was probablycloser to the surface, the twofold coordinated oxygenatoms are distinguished as independent features [220]. In correspondence to the rutile (1 1 0) surface, one might expect that the twofold coordinated oxygen atoms are removed easilyupon annealing in UHV, and thus give rise to point defects. Several imperfections with atomic dimensions are identi®ed in the atomic-resolution image in Fig. 29. At this point it is not clear which one of these features, if any, are indicative of oxygen vacancies. Their number densityis verysmall, certainlysmaller than the usuallyquoted 5±10% [116,128] on rutile (1 1 0). This would be in agreement with the (calculated) low surface energyof the (1 0 1) surface [70]. Calculations of the electrostatic potential byWoning and van Santen [223] also predict that the rutile (1 1 0) surface can be reduced easier than the anatase (1 0 1) surface. 102 U. Diebold / Surface Science Reports 48 (2003) 53±229

Ê Fig. 29. STM image (Vs 1:22 V, It 1:23 nA, 130 Ð Â 60 A) of an anatase (1 0 1) surface. Four features could possibly be representative of oxygen vacancies; single black spots (A), double black spots (B), bright spots (C), and half black spots (D). The densityof these atomic defects is rather small, con®rming the theoreticallypredicted high stabilityof anatase (1 0 1). From Hebenstreit et al. [220]. # 2000 The American Physical Society.

2.6.2. Anatase (0 0 1) The stable, autocompensated anatase (0 0 1) surface exhibits exclusively®vefold coordinated Ti atoms, as well as twofold and threefold coordinated oxygen atoms, see Fig. 30a. Calculations show that the corrugation increases somewhat upon relaxation, from 0.82 to 0.92 AÊ [70]. The most detailed structural investigations on this surface so far have been performed on thin ®lms, grown epitaxiallyon the SrTiO 3(0 0 1) substrate [218,219,221]. The SrTiO3(0 0 1) surface shows a very good lattice match with the anatase (0 0 1) surface (À3%), but a poor one with the rutile phase. Their anionic sublattices bear substantial resemblance despite their overall crystallographic dissimilarities

[224]. Heteroepitaxial growth of TiO2 can be regarded as a continuous formation and extension of the oxygen atom network from the SrTiO3 substrate into the ®lm. Within this oxygen sublattice the (relativelysmall) Ti cations arrange in their appropriate sites. Formation of interfaces where the oxygen sublattice continues and the metal cation sublattice changes abruptlyis often exploited for thin ®lm heteroepitaxyof metal oxides [225]. On SrTiO3(0 0 1) epitaxial, stoichiometric anatase thin ®lms of high crystallinequalityhave been grown byseveral techniques [224,226] and are stable up to 1000 8C [221]. The (0 0 1)-(1 Â 1) surface is not verystable, however, and reconstructs when heated to elevated temperatures [219,227±229]. Onlyon as-grown samples that were slightlycontaminated with carbon a (1 Â 1)-terminated surface was observed [218]. Herman et al. [219] were the ®rst to point out that a two-domain (1 Â 4) reconstruction formed on an anatase ®lm on SrTiO3 after sputtering and annealing the (1 Â 1) surface in UHV. Based on angle-resolved mass-spectroscopyof recoiled ions (AR-MRSI) a `microfaceted' model was proposed. In this model (1 0 3) facets are exposed which contain twofold oxygen and both fourfold and ®vefold coordinated Ti atoms. Such a model resembles in many ways the microfacet model for the rutile (1 0 0)-(1 Â 3) surface discussed in Section 2.3.2. The appearance of the (1 Â 4) reconstruction in STM [228] is not consistent with the (1 0 3) microfaceted model. Based on the STM images an `added and missing row model' was proposed [228]. Based on ®rst-principles calculations, Lazzeri and Selloni suggested the so-called `added molecule (ADM)' structure, see Fig. 30b and c [230]. The high-resolution STM and NC-AFM images in Fig. 31 are consistent with the ADM model [229]. Kinks and defects in the bright rows of the (1 Â 4) structure as well as the faint lines between the bright lines (which would be at the location of the Ti(5) atoms) ®t verywell to the proposed model. The NC-AFM images show elevated features which are also consistent with these added features. U. Diebold / Surface Science Reports 48 (2003) 53±229 103

Fig. 30. (a) Relaxed (0 0 1)-(1 Â 1) surface of anatase TiO2. (b) Relaxed structure of the `ad-molecule' (ADM) model for the 1 Â 4 reconstruction. (c) Projection of the atomic positions of the ADM model on the plane perpendicular to the y direction. Dots with different sizes represent atoms belonging to different planes parallel to the ®gure. Dotted lines represent bonds in the ideallybulk-truncated surface. The length in AÊ ngstrom of some surface bonds is indicated. a is the theoretical in-plane bulk lattice spacing (a 3:786 AÊ ). x and y correspond to the [0 0 1] and [0 1 0] directions. From Lazzeri and Selloni [230]. # 2001 The American Physical Society.

2.6.3. Other anatase surfaces As seen from Table 4, the anatase (1 0 0) surface should be quite stable. While this plane is not a terminating face at an equilibrium-shape crystal (Fig. 27), such planes are observed in powder materials. Unpublished STM results from this author's group [231] show that the (1 0 0) surface forms a(1Â n) reconstruction, which can be explained bya (1 0 1)-microfaceted model. The (1 0 3) surface also reconstructs to a (1 Â n) termination, but a persistent Ca contamination on this crystal was always present and excludes a de®nitive statement about the structure of this surface.

2.7. Conclusion

Much has been learned about the surface structure of the titanium dioxide system in recent years. Two somewhat contradictorylessons can be drawn from all this work: (1) simple approaches work well for obtaining a ®rst guess of surface structures and (2) oxide surfaces are even more complicated than anticipated. 104 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 31. SPM images of the anatase (0 0 1)-( 1 Â 4) surface. (a) Wide range STM image following annealing in oxygen and in UHV. Reconstructed step edges run along h001i directions, black arrows mark rows that contain a kink, and white arrows mark positions of anti-phase boundaries (APB). (b), (c) High-resolution STM shows that the bright rows are composed of two parallel rows of bright spots and defects. Between the rows are two fainter rows. (d) Between the bright rows are three fainter rows where a local (1 Â 5) periodicityexists. (e) High-resolution NC-AFM image. (f) Line scans from the image in (b) taken along a line perpendicular to the bright rows (A±A) and parallel to the bright rows (B±B). A tunneling current set point of 1.0 nA was used in (a)±(d). (a) Vs 1:0 V. From Tanner et al. [229]. # 2002 The American Chemical Society. U. Diebold / Surface Science Reports 48 (2003) 53±229 105

It is comforting that the elemental rules for predicting surface terminations outlined in Section 2.2.1.1 work so well for predicting the structure of (1 Â 1) terraces and step edges of all the orientations of both rutile and anatase surfaces. The extensive theoretical work has helped to re®ne the understanding of surface relaxations, and the level of detail on the atomic geometryof the TiO 2(1 1 0) surface is certainly comparable to that of certain elemental semiconductors or metals. On the other hand, scanning probe techniques have unraveled a veryrich picture of surface structures. One interesting theme is the interplaybetween surface structure and bulk defects in rutile; the reduction state of the crystal is quite important for the presence of different structural features under exactly the same preparation conditions. It remains to be seen if such a behavior is also present for other metal oxides, or even in TiO2 anatase. Total-energycalculations have helped enormouslyto con®rm models for surface reconstructions, and have acted as a warning sign when models derived from experimental results were too naõÈve. The most credible models often are not simple variations of the TiO2 bulk structure, see for example, the added `Ti2O3'-row reconstruction of rutile (1 1 0)-(1 Â 2) and the ADM model of the anatase (0 0 1)-(1 Â 4) surfaces.

The expanding data base has made rutile TiO2 a verypopular model systemfor metal oxides. Nevertheless, there are still manyopen questions concerning the crystalstructure of rutile surfaces as pointed out throughout this section. One interesting aspect is the advent of surface studies on anatase. From the recent progress in synthesizing single-crystalline anatase samples with high purity as well as the successful growth of epitaxial thin-®lms on appropriate substrates, one should expect an increase in the interest in anatase in the near future.

3. Electronic and vibrational structure of TiO2 surfaces

An excellent introduction to the (bulk) electronic structure of transition metal oxides was given by Cox [232]. This was followed up with a detailed discussion of the surface electronic structure in the book on oxide surfaces byHenrich and Cox [1]. Since publication of this book in 1994, much progress has been made in the theoretical understanding of TiO2 surfaces. Increasinglypowerful computational approaches have been used as is described in a large number of recent publications [68,75,

100,101,191,233±249]. DFT calculations have helped in understanding the structure of TiO2 surfaces, and have been a ®rst warning sign when structural models based on experimental observations were too simplistic (e.g. the missing-row reconstruction for the (1 1 0) surface [233,248] and the microfacet model for the (1 0 0) surface [179,250], see Section 2). The current trend is to calculate and understand adsorption of molecules and metals, and the pertaining literature is reviewed in Sections 4 and 5. The basic understanding of the electronic and vibrational structure of clean TiO2 surfaces as given in [1] is still valid; consequently, this part of this review is held brief. In the following a distinction is made between stoichiometric surfaces, and oxygen-de®cient, `reduced' ones. This distinction is made somewhat arbitrarilyas defects are introduced rather easilyin rutile, and almost everywork that discusses clean TiO2 surfaces is also concerned with reduced, defective surfaces.

3.1. Stoichiometric TiO2 surfaces

The electronic structure of TiO2 has been calculated using a wide varietyof theoretical approaches with varying degree of sophistication [68,100,101,191,233±248,251±258]. There is wide agreement 106 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 32. Mulliken projected densities of states of a three-layer slab of a stoichiometric TiO2(1 1 0) surface. From Paxton and ThieÃn-Nga [245]. # 1998 The American Physical Society. that the surface electronic structure is not too different from that of the bulk. No surface states are observed or predicted, except for non-stoichiometric surfaces (see Section 4). The occupied states are mostlyO2p derived, but exhibit a signi®cant degree of covalency,see Fig. 32. The use of LDA vs. GGA in DFT, and the inclusion of spin-polarization, has been shown to cause little change in the overall features of the LDOS of stoichiometric surfaces [245].The hybridization of the Ti levels with oxygen are experimentally determined with resonant photoemission [259±262]. When the photon energyis swept across the Ti3p absorption edge, the photoemission cross-section for Ti3d-derived states increases [137]. These resonances can be used for a qualitative estimate of Ti±O hybridization. Projected partial density of states have also been extracted from photoelectron diffraction measurements [263]. The iono-covalent character of the

TiO2(1 1 0) surface and of TinOm clusters with different sizes and charges has been studied by Albaretetal.[247]. On average, the Ti charge is close to 1.7 (as compared to the formal oxidation state of 4), and the oxygen charge is close to À0.85 (formal oxidation state of À2), with onlysmall variations depending on the system. On the TiO2(1 1 0) surface, the covalencyof bonds between the bridging oxygen atom and the underlying sixfold coordinated Ti atoms is enhanced compared to other surface bonds. This is similar to a Mullikan charge analysis of the calculations by Paxton and ThieÃn-Nga [245]. The conduction band is mostlyTi3d-derived. The octahedral coordination causes a crystal-®eld splitting of the d orbitals into two sub-bands, see Fig. 33. The eg orbitals (dz2 and dx2Ày2 ) point U. Diebold / Surface Science Reports 48 (2003) 53±229 107

Fig. 33. X-rayabsorption spectra of the Ti2p edge of a stoichiometric, well-ordered rutile TiO 2(1 1 0) surface. The effect of the crystal ®eld of the oxygen ligands on the 3d electrons is sketched.

directlytoward the oxygen ligands forming s-typeorbitals. The t 2g orbitals (dxy,dxz and dyz)pointin- between the oxygen neighbors and form p-type bonds. The peaks in inverse photoemission spectra [264±266] have been interpreted along these lines. The crystal-®eld splitting is clearly seen in X-ray absorption, see Fig. 33. Naturally, these states are extremely sensitive to the presence of point defects and other imperfections [265]. Because the crystal ®eld splitting changes when the con®guration of the oxygen ligands is slightly altered in different structures (note the differences in bond angles and bond lengths in rutile vs. anatase, Fig. 2), X-rayabsorption spectra are quite sensitive to the crystal structure and to local imperfections [267±269]. Their shape can successfully be calculated with short-range models [269,270]. XAS is also quite useful to investigate the oxidation/reduction reactions that occur when reactive metals are deposited on TiO2 [271,272],see Section 4.

While calculations of the electronic structure of TiO2 abound, this author is aware of onlyone photoemission experiment where measured dispersions were related to the calculated band structure

[273], see Fig. 34. Normal-emission spectra were taken on a TiO2(1 1 0) and a TiO2(1 0 0) samples. Two angles of incidence for the photon beam were used. Because synchrotron radiation is polarized, this allowed the identi®cation of the symmetry of the valence band states based on selection rules. Such an assignment is critical, because the valence band in photoemission is broad and complex (Fig. 35). Photon energies below the Ti3p to 3d resonance have been used in the analysis. Very ¯at bands have been found in reasonable agreement with theory. 108 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 34. Calculated and experimental dispersions for rutile TiO2 along the X±G±M directions of the reduced zone. The data were obtained from normal-emission spectroscopyfrom TiO 2(1 1 0) and TiO2(1 0 0) surfaces. The ®nal states bands for hn 25 eV are also shown. Closed and open symbols represent emission from different Brillouin zones in the extended zone scheme. Experimental bands are labeled bythe symmetryderived from dipole selection rules. From Hardman et al. [273]. # 1994 The American Physical Society.

Fig. 35. (a) Photoemission spectra (hn 35 eV, normal emission) from the valence band region of a sputtered and UHV- annealed, clean TiO2(1 1 0) surface. After adsorption of molecular oxygen at room temperature, the defect state in the band gap region disappears and the spectrum shifts by0.2±0.3 eV to higher binding energydue to band bending. A Shirley background was subtracted from both spectra. (b) Schematic diagram of the band-bending effect due to donor-like surface defect states. Surface oxygen vacancies create a defect state and electrons are donated to the system. A charge accumulation layer is created in the near-surface region and the bands in the n-type semiconducting TiO2 sample bend downwards. U. Diebold / Surface Science Reports 48 (2003) 53±229 109

3.2. Reduced TiO2 surfaces

3.2.1. Defect states As discussed in Section 2.2.1.4, annealing at high temperatures (or bombarding with electrons) reduces the TiO2(1 1 0) surface and creates point defects in the rows of bridging oxygen atoms (see Fig. 14). Fig. 35 shows two typical photoemission spectra from the valence band region which exemplifythe presence of oxygenvacancies. The solid line is from a (blue) TiO 2(1 1 0) crystal after sputtering and annealing in UHV. The defect state in the band gap is clearlyvisible. It shows almost no dispersion when the emission angle is changed [274]. Upon exposure to molecular oxygen gas at room 18 temperature, the defect state disappears. As was shown with isotopicallylabeled O2 experiments [127,275], the gaseous oxygen dissociates and ®lls the vacancies, which quenches the defect state. (See

Section 5.1.3 for more detail on oxygen adsorption on defective TiO2(1 1 0) surfaces.) In resonant photoemission the defect state shows a behavior clearlyindicative of a Ti3d-derived nature [259,261,262]. The presence of point defects is also visible in other spectroscopies, e.g. a peak at 0.75 eV in electron energyloss spectra [276,277], and a shoulder in XPS which is formallyassigned to a Ti 3 oxidation state (Fig. 36). The defect state in the band gap is usuallynot reproduced in theoretical calculations, however. One possible reason was pointed out byLindan et al. [233]. Spin-polarized DFT calculations of `reduced' models system (where all the bridging oxygen atoms were removed, and the coordinates were relaxed) show localized band gap states formed byTi3d orbitals, see Fig. 37 [233,245]. In spin- paired calculations, this feature is generallynot present in the band gap [248]. The nature of the defect state is easilyexplained [1,233]. Removal of a neutral oxygen atom leaves behind two electrons which previouslyoccupied O2p levels in the valence band. These states are no longer available, and the electrons must go into the conduction band, the bottom of which is formed byTi3d states. Both, the neighboring ®vefold and the sixfold Ti atoms receive an electron, and these electrons are unpaired [233,245]. Although the defect state is 100% spin-polarized, the interaction between spins is too weak for magnetic ordering to occur. In the HF approximation, the energydifference between ferromagnetic and

Fig. 36. XPS from a TiO2(1 1 0) (a) stoichiometric surface (b) after annealing in UHV, with point defects and (c) after sputtering. 110 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 37. Densities of states g(e) for spin-up and spin-down electrons calculated for a stoichiometric TiO2(1 1 0) surface after structural relaxations. From Lindan et al. [233]. # 1997 The American Physical Society. antiferromagnetic alignment is calculated to be 0.1 meV only [278]. To this author's knowledge, the spin- polarized nature of the 3d state on defective TiO2(1 1 0) surface has not been veri®ed experimentally.

3.2.2. Band bending Reduced titanium dioxide is an n-type semiconductor, and band-bending effects accompany the adsorption of gases or metals. An example for such a band-bending effect is clearlyvisible in Fig. 35. When oxygen vacancies are present, the extra electrons in the vacancies act as donor-like states that create an accumulation layer in the near-surface region. This causes a downward band bending. After adsorption of oxygen, only minimal changes occur in shape of the valence band in Fig. 35, but there is a rigid shift of all peaks in the photoemission spectrum upwards by0.2±0.3 eV. This is caused bya downwards shift of the Fermi level and an `unbending' of the bands.

3.2.3. Identi®cation of the reduction state with spectroscopic techniques The reduction state of the surface, i.e., the presence of lower oxidation states, can easilybe identi®ed with XPS and a varietyof other spectroscopic techniques [276]. Fig. 36a shows the Ti2p region of a clean, stoichiometric TiO2 surface. In spectra with a carefullycalibrated energyscale, the Ti2p 3/2 peak, attributed to (formally) Ti4 ions, is located at 459.3 eV, and the O1s peak at 530.4 eV [279].Note, however, that surface defects can cause band bending and a rigid shift of the whole spectrum, as discussed above. This probablyaccounts for the scatter in experimental peak positions. A mean of 458.7 eV of 16 literature data is given in [280], and a similar value (458.5 eV) is sometimes used as an `internal standard'

[281] in XPS experiments. The Ti2p spectra of TiO2 and a varietyof other Ti compounds have been calculated with a single impuritycluster model, and the experimental spectra (including the satellite features at higher binding energy, not shown in Fig. 36) were reproduced quite well [282].

After annealing a stoichiometric TiO2 surface in UHV (and creating point defects), a shoulder in the Ti2p spectrum appears, Fig. 36. This shoulder is attributed to a formal Ti oxidation state of 3. U. Diebold / Surface Science Reports 48 (2003) 53±229 111

Fig. 38. Raw HREELS data (bottom) of a clean TiO2(1 1 0) surface show losses caused bymultiple excitations of Fuchs± Kliewer phonons. These can largelybe removed from the spectrum bya Fourier deconvolution procedure. From Henderson [128]. # 1996 Elsevier.

Sputtering preferentiallyremoves oxygenfrom the surface and creates lower oxidation states which broadens the spectrum further (Fig. 36). The peak shifts associated with these lower oxidation state have been obtained bypeak ®tting byseveral authors, and a review of different values is given in [280].

3.3. Vibrational structure

The studyof the lattice dynamicsof TiO 2 with charged particles is complicated bythe high cross- section for excitation of optical phonons. HREELS measurements are dominated bythe losses from these Fuchs±Kliewer phonons [128,283±286], see Fig. 38. The raw spectrum in the lower panel in Fig. 38 shows the primaryphonon losses (at 365, 445 and 755 cm À1) and up to four multiple scattering events which give rise to the higher lying peaks. Because the electric ®eld extends deep into the bulk, these phonons carrylittle information about the state of the surface, although theydo change position and intensityupon sputtering [283±286] or metal deposition [280]. Unfortunately, these high-intensity peaks often overshadow small features which would be indicative of adsorbates on the sample. This is a general characteristic of all ionic oxides, and two techniques have been employed in order to circumvent this problem and make HREELS accessible to adsorption studies. Wu et al. [287] used a high energyof the primaryelectron beam and took HREELS spectra in an off-specular direction on a 112 U. Diebold / Surface Science Reports 48 (2003) 53±229

MgO(0 0 1) crystal. A more elegant way is to Fourier deconvolute the spectrum as suggested by Cox and Williams [288]. While this does not affect the primarylosses, it clears the spectrum of most of the intensityat higher energies, as shown in the upper panel of Fig. 38. (The origin of the remaining peak at 1515 cmÀ1 is discussed in [128].)

4. Growth of metal and metal oxide overlayers on TiO2

Several excellent reviews on the growth of metals on oxide substrates are given in [4,8,10,63].A comprehensive discussion of the structural, electronic, and chemisorption properties of metals on several metal oxides, including TiO2, was given byCampbell [4]. Metal overlayer growth speci®cally on TiO2(1 1 0) has been discussed bythis author in [63], and more recentlybyPersaud and Madey [10]. The heteroepitaxyof metal oxides on metal oxide substrates was reviewed byChambers [289] in a recent paper.

This ®eld of the surface science of TiO2 is veryactive, with a recent emergence of theoretical calculations of adsorbate studies on metals (and metal oxides) on TiO2. Fig. 39 shows the elements that have been studied on single-crystalline TiO2 surfaces, and Table 6 gives a short summaryof the main results. The growth mode, interfacial reaction, structure, and thermal stabilityof metal overlayerson

TiO2 follow clear trends across the periodic table, with the oxygen af®nity of the metal overlayer being one of the most decisive factors for the properties of the metal/TiO2 systems. These trends are discussed in the ®rst part of this section. Details about different systems, with an emphasis on the literature after 1997 are then discussed in the next part.

4.1. Overview and trends

4.1.1. Interfacial reactions

It has been pointed out in previous reviews of metal overlayer growth on single-crystalline TiO2 surfaces [4,10,63] that the metal overlayers' reactivity towards oxygen is a very important parameter for

Fig. 39. Adsorption and/or growth of the shaded elements on single-crystalline TiO2 surfaces was studied and are described in the text and is summarized in Table 6. U. Diebold / Surface Science Reports 48 (2003) 53±229 113 predicting a varietyof properties of the metal/TiO 2 interface. The more recent papers published since these reviews were written are consistent with this trend. Thermodynamic considerations for metal ®lms on oxides were given by Campbell [4] and, for the speci®c case of metals on TiO2 byPersaud and Madey [10]. Generally, if a metal M is deposited onto TiO2, then M should reduce the substrate and itself become oxidized if the reaction

M TiO2 ! MOx TiO 2Àx (1) is thermodynamically favorable, i.e., if the change in the standard free energy, DHf is negative. In the case of a TiO2 substrate, formation of lower oxidation states such as Ti2O3 or TiO is possible (see Fig. 4). Thus, the DHf per mole of oxygen of oxide formation should be compared with 2TiO ! 2Ti O s1 O g DH DH À DH 364 kJ=mol (2) 2 2 3 2 2 f f;Ti2O3 f;TiO2 or Ti O ! 2TiO 1 O g DH DH À DH 483 kJ=mol (3) 2 3 2 2 f f;Ti2O3 f;TiO2 in order to judge whether or not such a reaction will take place. Thus, in thermodynamic equilibrium the reaction (Eq. (1)) can take place in principle if DHf is more negative in Eq. (1) than in Eq. (2) or Eq. (3). The heats of oxide formation were compiled byCampbell [4] and are reproduced in Table 5. (Note that the numbers in Table 5 are per mole of oxygen, while others [10] have referenced formation enthalpies per mole of molecular O2.) The heats of oxide formation are high for transition metals on the left side of the periodic table, and when a reactive metal is vapor-deposited on TiO2 at room temperature, the corresponding oxidation/ reduction reaction results in a reacted interface. This is clearlyseen in XPS measurements of verythin overlayers, see Fig. 40. (The ®lm thicknesses in the ®gure refers to quartz-crystal microbalance readings; i.e., theyrepresent the nominal thickness if the overlayerwere spread out uniformlyacross the

Table 5 Heat of formations of the most stable oxide of the metala

b Heats of formation of oxide (DHf in kJ/mol O) Metal >0 Au 0toÀ50 Ag, Pt À50 to À100 Pd À100 to À150 Rh À150 to À200 Ru, Cu À200 to À250 Re, Co, Ni, Pb À250 to À300 Fe, Mo, Sn, Ge, W À300 to À350 Rb, Cs, Zn À350 to À400 K, Cr, Nb, Mn À400 to À450 Na, V À450 to À500 Si À500 to À550 Ti, U, Ba, Zr À550 to À600 Al, Sr, Hf, La, Ce À600 to À650 Sm, Mg, Th, Ca, Sc, Y a After [4]. b For the most stable oxide of the metal. 114 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 40. The Ti2p region of several metal overlayers before (dotted lines) and after (full lines) overlayer deposition at room temperature. The sharp peaks of the clean surfaces are indicative of Ti atoms on a stoichiometric surface with a (formally) 4 Ti oxidation state. The more reactive the metal overlayer, the more the peaks smear out, indicating a reduction of the TiO2 substrate (see also Fig. 36). From [63]. surface. However, this growth morphologyrarelyoccurs, see below.) Fig. 40 shows the Ti2p XPS levels from the TiO2 substrate before and after deposition of four different metals, representative for very different levels of reactivity. As discussed above (Section 3.2 and Fig. 36), the Ti2p core levels are a verygood indicator for the stoichiometryof TiO 2 surfaces and the appearance of lower oxidation states (the shoulders at the lower binding energyside in Fig. 40) indicate a reduction of the substrate. The strength of this reduction reaction scales with the reactivityof the overlayer. It is virtuallyabsent for Cu and strongest for Hf, which exhibits an oxide heat of formation of more than 550 kJ/mol, see Table 5. XPS core level measurements of the corresponding overlayer metals show that the substrate reduction is accompanied byoxidation of the overlayer, in agreement with Eq. (1). Such a solid-state oxidation/ reduction reaction was observed for Co but not for Ni, which puts the limit around an oxide heat of formation of 250 kJ/mol, see Table 5. (Because the numbers in Eqs. (2) and (3) are referenced to bulk values, this smaller value derived from surface experiments is not too disturbing.) Hence the borderline between metals that induce an oxidation/reduction reaction, and the ones that do not, lies roughlyalong an axis connecting Co and Re in the periodic table (Fig. 39). Do the observed changes in XPS line shape indicate a mere electron transfer from the reactive overlayer to the substrate Ti ions, or is oxygen physically extracted from TiO2 and incorporated in the overlayer as suggested in Eq. (2)? At least for the more reactive metal overlayers, the latter is clearly the case. The impressive transmission electron microscopyimage of a Nb/TiO 2(1 1 0) interface, discussed below (Section 4.2.11), clearlyproves this point. A sharp, but reacted interface is seen. ComplementaryEELS measurements, taken across the interfacial region, support the presence of a reduced TiO2 layer. Formation of a disturbed interface is consistent with surface studies of other reactive metals, e.g. for Hf U. Diebold / Surface Science Reports 48 (2003) 53±229 115

(see Section 4.2.8). The LEED pattern disappears rapidlywith coverage, indicating interfacial disorder, and the oxidation/reduction reaction does not stop until at least 2 ML are deposited, see Fig. 40.This clearlymeans that oxygen is extracted from the substrate and incorporated into the ®lm. Interestingly,in all the metal overlayers studied, reactive and non-reactive ones, a sharp interface is formed between the overlayer metal and the Ti when the ®lm is grown at room temperature. No intermixing for Ti and the overlayer metal was observed, even in the cases where stable Ti alloys would exist, e.g. for Pd or Al overlayers. The oxidation/reduction reactions at the interface lead to a substantial re-arrangement of atomic positions and/or charge that ultimatelycauses the effects observed in Fig. 40. Ionization of the overlayer metal introduces band gap states (see for example, the discussion on Fe adsorption below (Section 4.2.17)). A similar charge transfer and formation of gap states has been observed for all alkali metals. Donation of charge leads to changes in band bending and work function for all the metal overlayers. This is as discussed in great detail in [4].

Unfortunately, very few theoretical calculations of reactive metal adsorption on TiO2 exist at this point. ThieÃn-Nga and Paxton [290] have made a comparison across the 5d transition metals. These authors have found covalent bonding and verylittle charge transfer. However, this studytook into account onlyone adsorption site (on top of ®vefold coordinated Ti atoms) and no relaxations, in clear contradiction with the experimentallyobserved, reacted interface. There is no spectroscopic evidence for an oxidation/reduction reaction for metals equallyor less reactive towards oxygen than Ni. Core levels shift somewhat because of band bending or ®nite size effects in the small clusters formed bythe metal overlayer. However, the substrate core levels show no sign for the formation of lower oxidation states, as seen in Fig. 40 for Cu. The fact that no interfacial oxidation/ reaction occurs for unreactive metals does not preclude relaxations to happen at the interface. In contrary, the onlysystem where interfacial coordinates were measured (an X-raydiffraction investigation of the

Cu/TiO2(1 1 0) system by Charlton et al. [107]) shows substantial relaxations of the substrate atoms.

4.1.2. Growth morphology (thermodynamic equilibrium) It is useful (though a quite rough oversimpli®cation, see below) to distinguish three ®lm growth modes in thermodynamic equilibrium. When the difference between the surface free energy of the clean substrate, gsubstrate, and surface free energyof the overlayermetal, gmetal, is greater than the interfacial energy, ginterface, i.e.

ginterface > gsubstrate À gmetal (4) cluster growth (also called Volmer±Weber growth) should take place. When it is less, the ®lm should wet. For thicker ®lms, growth can proceed in a layer-by-layer fashion (Frank±van der Merwe growth mode). Often, epitaxial strain increases in thicker ®lms and breaks up the overlayer (layer clusters or Stranski±Krastanov growth mode). Because the surface energyof virtuallyall clean metals is higher 2 than that of TiO2 (where an experimental value of 0.35 J/m was reported [291]) the term on the left side is negative. Hence, cluster growth should occur, unless ginterface itself has a negative value. It has been suggested [10,63,292] that this is the case for veryreactive overlayers,where an interfacial reaction as in Eqs. (1)±(3) is thermodynamically favored. Consequently, the tendency to wet the substrate should correlate with the numbers given in Table 5. Fig. 41 shows that this is indeed the case, at least to ®rst approximation. The growth morphologies at the right side of the ®gure were extrapolated from combined XPS and LEIS measurements for four 116 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 41. Trends for the attenuation of the Ti LEIS signals for four selected overlayers on TiO2(1 1 0) and schematic drawing of the initial stages of ®lm growth for Cu (3D clusters), Fe (¯at islands), Cr (2D islands followed by3D growth) and Hf (formation of a continuous overlayer). From [63]. selected metal overlayers (Cu, Fe, Cr and Hf; the same ones as in Fig. 40). Low-energyion scattering is particularlyuseful for growth studies of ultrathin ®lms. It is primarilysensitive to the top-most surface layer. If the ®lm wets and grows in a 2D fashion, the LEIS signal from the substrate should linearly decrease with coverage and approach zero upon completion of the ®rst monolayer. This is the case for Hf, a metal with a veryhigh heat of oxide formation. In the case of 3D growth, the substrate signal should be visible up to fairlyhigh coverages, as is the case for Cu. The strength of the oxidation/ reaction, as evidenced in Fig. 40, is directlycorrelated with the growth mode extrapolated from the LEIS measurements. As XPS measurements have been performed for most of the metal overlayers considered to date (see second column in Table 6), a prediction of the growth morphologycan be made for most overlayer metals. For very unreactive overlayers, the work of adhesion, i.e., the work required to separate the overlayer from the substrate

Wadh gsubstrate gmetal À ginterface (5) can directlybe extrapolated from contact angle measurements on a microscopic scale [293]. This was done in an impressive waybyCosandeyand Madey [294] for Au clusters on a TiO2(1 1 0) surface, see below. According to Table 5, gold does not react with oxygen, which is consistent with the large contact angle observed. (Ironically, such supported small gold clusters do promote the catalytic oxidation of CO, see below!.) The correlation between heat of oxide formation, interfacial reaction and growth mode is intriguingly simple, but neglects several important issues. The growth modes discussed above are strictlyonlyvalid at thermodynamic equilibrium, whereas ®lm growth is often performed at low temperatures, e.g. convenientlyat or near room temperature. The substrate is considered rigid and its structure and Table 6 Summary of growth results of metal and metal oxide overlayers on single-crystalline TiO2 surfaces Metal/substrate Interfacial reaction Techniques Growth Thermal stabilityReference morphology/structure Li/rutile (1 1 0), Formation of Ti3 Semi-empirical Equilibrium position Intercalation in [314±316,319] anatase (1 0 1) states in the band gap Hartree±Fock between two bridging anatase much easier calculations, ESD oxygen on rutile (1 1 0), than rutile in structural void in anatase (1 0 1) Na/rutile (1 1 0)- Reduced TiOx, Ti3d XPS, AES, NEXAFS, Fig. 42 [160,205,304,305, (1 Â 1), rutile gap states promotes EELS, resoPE, LEED, 320±332] 53±229 (2003) 48 Reports Science Surface / Diebold U. (1 1 0)-(1 Â 2), rutile adsorption of CO2 D scattering, STM, (4 4 1), CO2, NO UPS, Hartree±Fock, molecular dynamics calculations K/rutile (1 1 0), rutile Reduced TiOx, ®rst Ellipsometry, TPD, Metallic > 1 ML; Desorbs as KOx, [278,319,322,333±342] (1 0 0), O2,CO2 ML oxidized UPS, IPS, XPS, D multilayers at 140 K reduces the substrate scattering, ARUPS, ARXPS, LEED, RHEED, SEXAFS, HREELS, MIES, UPS Hartree±Fock, LCAO Cs/rutile (1 1 0), rutile Reduced Ti, D scattering, MIES, First layer bound >0.5 ML desorbs at [321,342±344] (1 0 0)-(1 Â 1), complete ionization XPS, TPD, UPS, strongly, multilayers elevated temperatures CO2 for small coverages work function at low temperatures Ca/rutile (1 1 0) Segregating impurities; LEED, XPS, LEIS, Ordered structures Segregates towards [145±148,733] deposition from solution STM, atomistic the surface calculations Al/rutile (1 1 0), Reduced Ti at the XPS, UPS, AES, Fractional coverage Heating in UHV [281,345±349] rutile (100), interface, oxidized and LEED, STM disorders substrate; oxidizes Al and K, C metallic Al in the ®lm evenlydistributed reduced TiOx layer, clusters, no preferred no ordered structure sites Ê Ê 0 Ti and TiOx/rutile 4ATi produces 20 A EELS, LEIS, XPS, Ti clustering on top Bulk diffusion >700 K, [74,280,350] (1 1 0) of reduced layer SSIMS, conductivity, of reduced interface Ti main diffusing LEED species Hf/rutile (1 1 0) Reduced Ti layer, LEIS, XPS, SXPS, Hf 0 clustering on top Oxidation of Hf [63,351] oxidized Hf 4 at the LEED of oxidized Hf 4 /Hf x overlayer interface layer, see Fig. 43 117 118 Table 6 (Continued ) Metal/substrate Interfacial reaction Techniques Growth Thermal stabilityReference morphology/structure V/rutile (1 1 0), Ti3 , oxidized V, STM, AFM, UPS, At low coverages: binds Diffuses into substrate [12,14,15,352± rutile (1 1 0)- metallic for higher XPS, LEED, work on top of (1 Â 2) rows, >600 K 354,356,357,734,735] (1 Â 2), rutile coverages function, ARXPS, see Fig. 44, no LEED, (1 0 0) HREELS, AES, TPD ARXPS: adsorption site underneath bridging oxygens (after anneal at 473 K), 5 ML: epitaxial

bcc(1 0 0)[0 0 1]|| 53±229 (2003) 48 Reports Science Surface / Diebold U. (1 1 0)[0 0 1]TiO2 VOx /rutile (1 1 0), Ti remains oxidized SEXAFS, XPS, PES, Different vanadium >1100 K agglomeration [13,15,272,358±362] rutile (1 1 0)- work function, oxides dependent on of V2O3 particles (1 Â 2), methanol ARXPS, STM preparation conditions: VTiO3 at interface À6 V2O3:10 Torr O2 at room temperature, VO2: fractional ML anneal 473 K, VOx (x 1) annealing of V ®lm in UHV, 1 ML V2O3: methanol ! formaldehyde Nb/rutile (1 1 0), rutile Intermediate RHEED, AES, HRTEM, At room temperature: Formation of [58,93,190,363±376] (1 0 0), NbxTi1ÀxO2 reaction layer, EELS, STM, 2 nm thick interlayer, NbxTi1ÀxO2 solid mixed ®lms see Fig. 45 photoemission, MBE of 2 ML thick NbOx, solutions mixed ®lms, HF and metallic Nb on top, densityfunctional intermixing between calculations of doping Nb and TiO2 kineticallyhindered bcc(1 0 0)[0 0 1]|| (1 1 0)[0 0 1]TiO2 Cr/rutile (1 1 0) Reduced Ti, oxidized LEIS, XPS, LEED, Overlayer oxidized Diffusion into substrate [58,292,355,377±380] Cr at interface MEED, ARXPS, HF through a dynamic competes with formation calculations of doping exchange with lattice of metallic clusters at levels oxygen, metallic Cr on intermediate T; diffusion top bcc(1 0 0)[0 0 1]|| into bulk at high T (1 1 0)[0 01 ]TiO2 Mo/rutile (1 1 0) Ti3 and Ti2 AES, XPS, RHEED, ex Completion of three [382,736] at the interface situ AFM, adsorption monolayers followed by of Mo(CO)6 islands; different substrate pretreatments have no in¯uence on growth mode Mox/rutile (1 1 0) XAFS Well-dispersed Mo [383±392] oxides prepared by impregnation/calcination; preferentiallyMo dimers with Mo-Mo bond

parallel to 1 10 53±229 (2003) 48 Reports Science Surface / Diebold U. Mn/rutile (1 1 0) Reacted, disordered SXPS, XAS Oxidized Mn atoms, T > 625 8C: metallic [271] interface with reduced metallic Mn for Mn desorbs, formation Ti cations thicker layers of ternaryMnTiO x at the interface b-MnO2/rutile OPA-MBE, RHEED, 400±500 8C: [393] (1 1 0) LEED, XPS, XPD, pseudomorphic ®lms, AFM island growth, 500±600 8C: intermixing Fe/rutile (1 1 0), Ti3 and oxidized Fe; AES, LEED, LEIS, Flat clusters bcc(1 0 0) Encapsulation in UHV [17,262,355,378,379, rutile (0 0 1), Ti 3d-derived and Fe UPS, HREELS, [ 0 01]||(1 1 0)[0 0 1]TiO2 394±397] CO, O2 3d-derived band gap resonant photoemission, surface roughness states, see Fig. 46 XPS, MEED, ARXPS, in¯uences morphology, IPS oxygen ¯attens the ®lms creates FeOx and Fe2O3 phases Ru Ti substrate not Decomposition of Disordered nanoclusters Substitutional alloys [400,401] reduced Ru3(CO)12 at 300 8C, XPS, LEED, ARXPS RuO2/rutile (1 1 0) Intermixing at OPA-MBE, LEED, Stacks of ®lms with Extensive intermixing [289,400±402] the interface RHEED, XPS, XPD, decreasing degree of at 600 8C ARXPS, decomposition intermixing of Ru3(CO)12 Co/rutile (1 1 0) Minor oxidation/ XPS Assumed homogeneous Indiffusion at 500 K, [405] reduction reaction growth reduces support (SMSI), at 300 K less stable on reduced TiO2 Rh/rutile (1 1 0), No interfacial STM, AES, XPS, LEIS, Mainly(1 1 1) oriented Coalescence of particles, [407±409,411,413, rutile (1 1 0)- reaction [Rh(CO)6Cl]2 IRAS 3D particles, control encapsulates upon 737,738] (1 Â 2), CO, of particle size by heating (substrate and 119 rutile (0 0 1) seeding/growing treatment dependent) 120 Table 6 (Continued ) Metal/substrate Interfacial reaction Techniques Growth Thermal stabilityReference morphology/structure Ir/rutile (1 1 0)- STM, AES Round particles on Increase in particle [414,415] (1 Â 2), CO the (1 Â 2) rows, CO size upon annealing disrupts crystallites Ni/rutile (1 1 0), No interfacial reaction, UPS, XPS, STM, AES: S±K, STM: VW, Islands coalesce up [416±426] CO, air, interfacial Ni atoms HREELS, EXAFS, nucleation at step to 880 K rutile (1 0 0) slightlynegatively RHEED, LEED, SIMS, edges, `hut-clusters' charged DFT-embedded cluster with {1 1 1} and

{1 0 0} facets 53±229 (2003) 48 Reports Science Surface / Diebold U. Pd/rutile (1 1 0), No interfacial reaction STM, LEED, AES, Clusters fcc 111 In UHV: sintering, [19,162,172,297,303, CH3OOH, CO, CAICISS, RHEED, 1 20jj 110001Ti encapsulation. When 346,427±435,739,740] rutile (1 0 0)-(1 Â 3) HREELS, FT-RAIRS, possibly(1 1 0) or annealed in O2: encap- DFT-embedded cluster (1 0 0) at small sulation and spillover coverages Pt/rutile (1 1 0)- No interfacial reaction STM, LEIS, LEED, Volmer±Weber growth, Sintering, encapsulation [18,20,71,72,190,302, (1 Â 1), rutile ARXPS, MEED, XPS, nucleation at step edges on reduced TiO2 364,436±440,442±446, (1 1 0)-(1 Â 2), rutile AES, ESD, ®rst- on well-annealed substrates, see Fig. 47 741±746] (0 0 1)-(1 Â 3), principles calculations surfaces (1 1 1)|| CO, H2O (1 1 0)TiO2, (1 1 1)|| (1 0 0)TiO2: equal amounts of Pt 2 1 1 parallel and perpendicular to TiO2 0 10 Cu/rutile (1 1 0)- No interfacial reaction STM, LEIS, XPS, Clusters (see Fig. 48), High mobility, sintering [107,159,184,295,355, (1 Â 1), CO LEED, MEED, EELS, sharp interface [1 1 0] alreadyat low 364,377±380,396, AES, SXRD, UPS, (1 1 1)Cu||[0 0 1] temperatures 447±449,451,452, inverse PE, ARUPS, (1 1 0)TiO2 747±750] HRTEM, FT-RAIRS Ag/rutile (1 1 0)- No interfacial reaction XPS, LEIS, LEED, 3D cluster growth Sintering upon [308,346,453±455, (1 Â 1), rutile STM, surface annealing from 100 751,752] (1 1 0)-(1 Â 2), differential re¯ectance to 300 K O2 (high pressure) (SDR), AFM Au/rutile (1 1 0), No interfacial reaction [294] 2D initially, then 3D Sintering [294,346] (reviews) O2,CO (1 1 1) and (1 1 1 2) [22±24,306,307,309, oriented clusters, 310,461,464,465, see Fig. 49 753±758] U. Diebold / Surface Science Reports 48 (2003) 53±229 121 stoichiometryunaltered upon metal deposition. Atomicallysharp interfaces are assumed, while Hf, Nb, and other highlyreactive metals clearlyextract oxygenfrom the substrate, and some metals intermix with the TiO2 substrate at higher temperatures. Most importantly, any kinetic considerations are neglected, as well as the role of surface defects for nucleation and growth. Some of these issues are considered in the next section.

4.1.3. Growth kinetics, nucleation, and defects After decades of surface research and thousands of detailed ®lm growth studies, it is almost a platitude to saythat kinetics playa major role in the formation of overlayers.Verydetailed information about growth mechanisms can be extracted from statistical information such as island densityand height distributions, nucleation sites, etc. Interestingly, STM measurements on metal/TiO2 have not become available until fairlyrecently.Spatiallyresolved information of overlayerson TiO 2(1 1 0) have been obtained for veryfew reactive (in the sense described above) metals on TiO 2(1 1 0), as well as most of the unreactive overlayers (Rh, Ir, Ni, Pd, Pt, Cu, Ag, Au). However, very few growth studies have systematically varied growth parameters such as deposition ¯ux, temperature, or angle. Of the reactive metals, onlyAl and V have been investigated with STM. (The growth of alkalis is special and discussed below.) Because changes in the local electronic structure dominates STM images on TiO2, it is dif®cult to obtain atomic resolution from overlayers that are oxidized upon contact with the TiO2 substrate. The diffusion length of metal adatoms is small because the metal/substrate interaction is quite vigorous. Consequently, nucleation occurs at terraces, and steps and point defects are less important in the nucleation process. Spectroscopic measurements comparing ®lms grown on sputtered and smooth surfaces indicate some in¯uence of surface roughness, however. As outlined above, a wetting of the metal overlayer is expected for reactive metals from thermodynamic considerations. For example, in the case of V (see [12] and discussion below), the thicker ®lms cover most of the surface and become granular, consistent with Stranksi±Krastanov type growth mode. However, the growth does not proceed in the orderlylayer-by-layerfashion suggested in the sketch in Fig. 41. The non-wetting behavior of unreactive metal overlayers is clearly observed in STM images, see for example, Cu/TiO2(1 1 0)-(1 Â 1) below. In contrast to the more reactive metals, nucleation occurs preferablyat step edges. Interestingly,the size distribution of the 3D metal clusters is verynarrow, see the STM image of Cu/TiO2, below. A `self-limiting growth' was suggested to be responsible for this phenomenon [295]. The relativelyuniform cluster size makes such metal/TiO 2 systems good candidates for `¯at model catalysts' where the size±dispersion can be very well controlled. Even when a metal overlayer grows in a 3D fashion, quasi-2D islands are formed during the initial stages of the growth for kinetic reasons. These 2D islands are of particular interest; all the overlayer atoms have substrate atoms as neighbors and therefore possiblydifferent chemisorptive properties [4]. The 2D to 3D transition occurs at a critical coverage. It has been measured for the noble metals Cu, Ag, and Au to range somewhere around 0.2 ML. For a similar system, Cu=ZnO 1010, it has been shown that the 2D to 3D transition is also dependent on the number of nucleation sites on terraces, e.g. adsorbates or defects. At perfectly¯at and clean surfaces, 3D growth is observed for the lowest coverages [296]. Defects might be of similar importance on a TiO2(1 1 0) surface. Recent STM work byFrenken and co-workers [297] has pointed out the in¯uence of defects on the size distribution and densityof clusters for Pd/TiO 2(1 1 0). The surface structure of rutile (1 1 0)-(1 Â 1) is quite corrugated and the bridging oxygen atoms protrude from the surface. It is tempting to assume that such a surface should act as a template for the 122 U. Diebold / Surface Science Reports 48 (2003) 53±229 formation of one-dimensional (1D) islands, either in the troughs or along the bridging oxygen atoms. This has not been observed with a direct-imaging method, however. The reaction with facile bridging oxygen atoms at the (1 Â 1) surface is possiblytoo vigorous and disruptive. Non-reactive metals, on the other hand, are too mobile at room temperature and form clusters starting at the verysmallest coverages. No measurements have been performed with low-temperature STM, where this aggregation could be prevented. On the TiO2(1 1 0)-(1 Â 2) structure, preferential nucleation on the (1 Â 2) rows was reported for both, reactive overlayers (see Fig. 44) as well as non-reactive ones.

4.1.4. Film structure and epitaxial relationships Experimentally, special adsorption sites have been identi®ed only in the case of alkali metals.

Ordered superstructures are observed for fractional monolayer coverages, e.g. TiO2(1 1 0)-c(4 Â 2)Na and K/TiO2(1 0 0)-c(2 Â 2)K. The various models for superstructures of Na/TiO2(1 1 0) are discussed in the context of Fig. 42. Alkali metals tend to sit at positions that maximize the contact to oxygen atoms, see Fig. 42c. Growth of ordered epitaxial overlayers has been observed for most transition metals. Exceptions (in the sense that no ordered LEED pattern was observed) are the veryreactive materials Ti, Al and Hf. The bcc metals V, Nb, Cr, and Fe grow with their (0 0 1) face parallel to the surface, and the [1 0 0] orientation aligned with the oxygen rows of the TiO2(1 1 0) surface. Surprisingly, epitaxy is good despite the fact that there is a substantial re-arrangement of atoms at the interface. For the most part, the non-reactive materials Ni, Pd, Pt, Cu, Ag, and Au all form fcc overlayers. In most cases, (1 1 1) facets of the clusters are parallel to the substrate, but other orientations have been observed as well, see Table 6.

4.1.5. Thermal stability of metal overlayers on TiO2-SMSI The thermal behavior of the overlayer metal can roughly be classi®ed in several groups. Metals with a verylow cohesive energysuch as the alkalis and Mn desorb upon annealing. Heating a TiO 2 substrate that has been covered bya veryreactive metal (Al, Ti, Hf) induces oxidation of the overlayer, accompanied byeither oxidation or further reduction of the substrate. The thermodynamically favorable reaction (oxidation of the overlayer and reduction of the substrate) is kinetically hindered at room temperature. At elevated temperature, the activation barrier is overcome, and a more complete oxidation of the overlayer takes place. Metals with an intermediate strength of the interfacial reaction (V, Nb, Cr, Ru) interdiffuse with the substrate, although some clustering of the overlayer metal can occur as well (Cr). Finally, the non-reactive metals all show sintering upon mild annealing. Cluster migration and coalescence has been observed for Pd/TiO2(1 1 0) [297] and might be the mechanism for other non-reactive overlayers as well. The group VIII A metals Fe, Co, Rh, Ir, Ni, Pd, Pt all show a veryinteresting behavior upon annealing in a reducing atmosphere that has been discussed in the catalysis literature for several decades. The acronym SMSI has been termed by Tauster [298] to account for the changes in catalytic activitywhen catalysts,consisting of these metals supported on TiO 2 or other reducible oxides (TaO5, CeO2, NbO, etc.), are reduced at elevated temperature. Adsorption of H2 and CO is drasticallyreduced, but competitive hydrogenation vs. hydrogenolysis reactions are greatly favored in SMSI systems. For example, methane production from CO or CO2 and H2 is enhanced bythree orders of magnitude [16]. Hence the SMSI phenomenon allows to tailor the selectivityof a catalystand has caused wide-spread interest. High-resolution electron microscopy(HREM) clearlyshows that some sort of reduced U. Diebold / Surface Science Reports 48 (2003) 53±229 123

Fig. 42. Summaryof models of Na adsorption on TiO 2(1 1 0). (a) The Na2O model was initiallyproposed byOnishi et al. [205,304,305] to account for the c(4 Â 2) overlayer observed with LEED. (b) Recently, the same group performed STM measurements that seemed to indicate that Na is in fact adsorbed on Ti atoms and not at the bridging oxygen atoms [323].A coordination to oxygen atoms seems much more likely, however (c) Two different adsorption sites of Na atoms at bridging oxygen atoms were considered, adjacent to one, or in between two bridging oxygen atoms, respectively. Ion scattering measurements are consistent with the `adjacent' con®guration [327]. Snapshots of molecular dynamics calculations [329,330], 2À re-drawn in (d), show that most Na adatoms are in `between' positions. Two proposals for the con®guration of the CO3 complex that forms upon adsorption of CO2 on Na-precovered surfaces are pictured in (e) and (f). A linear con®guration (e) is consistent with STM measurements [323]. (f) NEXAFS experiments indicate a tilted position [328]. titanium oxide migrates onto the clusters [299,300]. Special Ti3 sites on this layer and/or at its interface with the metal cluster are thought to be responsible for the changed reactivity [298,301]. For a while there has been uncertaintyas to whether the same phenomenon occurs when SMSI metal

®lms are deposited on single-crystalline TiO2 and annealed under UHV conditions. It has been shown to take place for some materials, e.g. Fe [17]. In other cases, such as Pt, different groups came to 124 U. Diebold / Surface Science Reports 48 (2003) 53±229 opposite conclusions using the same experimental techniques. It is now clear that the reduction state of the TiO2 substrate plays a decisive role on whether or not encapsulation of the supported metal clusters occurs, hence the effect can be sample-dependent. Some excess Ti atoms or ions must be in the near- surface region in order for the encapsulation to occur. The structure of encapsulated Pt clusters have recentlybeen investigated bythis author's group [20,302]. Details are discussed in the section on Pt overlayers below. Similar results have been obtained recently on encapsulated Pd layers [303].

4.1.6. Chemisorption properties

One of the main motivations for studying metals on TiO2 is the desire to synthesize model systems for supported metal catalysts [293]. As listed in Table 6, adsorption of simple molecules has been investigated for a few metal overlayers or clusters on TiO2. A verynice example is the promotional effect of the TiO2(1 1 0)-c(4 Â 2)Na overlayer for the adsorption of CO2 [205,304,305]. The investigation of Au/TiO2 has received a recent surge of interest in an attempt to understand low- temperature oxidation reactions [306]. In this context, recent high-pressure and high-temperature STM studies of supported clusters, performed in Goodman and co-workers [307±310] on Au and Ag, and by Bowker and co-workers [19] on Pd, are a particularlyinteresting attempt to bridge both the `materials' and the `pressure gap'.

4.2. Metals and metal oxides on TiO2

4.2.1. Lithium

Promotion of TiO2 with Li is of technological importance in dye-sensitized solar cells [311],for energystorage in lithium intercalated nanostructured TiO 2, and in humidityand oxygen sensors [312]. Doping with Li also affects the selectivityof TiO 2 catalysts in the conversion of methane to C2- hydrocarbons [313]. The onlyexperimental surface studyconducted so far has been an ESD investigation on thin polycrystalline ®lms [314,315]. Semiempirical Hartree±Fock calculations on Li adsorbed on (and incorporated in) rutile and anatase TiO2 have been reported byStashans et al. [316]. The Li ion can occupyan interstitial position in one of the structural voids in the anatase crystal without big distortions in the crystal structure (see Fig. 2 for the anatase structure). For rutile, problems with self- consistencyand large distortions around the Li ion point towards an unstable con®guration. This is in agreement with experimental observations that Li can be intercalated with higher probabilityin anatase than rutile [209,317,318]. The equilibrium position on the rutile (1 1 0) surface is located between two bridging oxygen atoms [316,319]. On the anatase (1 0 1) surface, the Li ion moves into one of the structural voids, underneath the top surface layer. In all cases, a Li-induced local one-electron energy level appears in the gap between the upper valence band and the conduction band and could be attributed to Ti3 states. This is similar to what is observed for other metals with a high af®nitytowards oxygen.

4.2.2. Sodium

The adsorption of Na on TiO2(1 1 0) has been studied with experimental [160,205,304,305,320±328] as well as theoretical [319,329±332] techniques. No reports for the growth on other faces exist at this point. The presence of Na stronglyaffects the electronic structure of TiO 2, via oxidation of the adsorbed Na and reduction of the Ti atoms in the substrate. Changes in the electronic structure also re¯ected bya strong change in work function, a downward band bending (see Fig. 35), and in the neutralization and scattering of D ions [321,322]. Resonant photoemission experiments [324] show that the Na-induced U. Diebold / Surface Science Reports 48 (2003) 53±229 125 feature in the band gap has a resonance behavior indicative of a Ti3d-derived feature, similar to the `defect state' in Fig. 35. A mechanism for the adsorption process is proposed in which a long-distance electron transfer occurs from atomic Na towards a ®vefold coordinated surface titanium atom. This is followed byphysisorptionof the Na cation at the surface [330]. In calculations this electron transfer from the Na3s atomic orbital to the surface gives rise to a highlyspin-polarized state in the band gap which is localized at a 5-fold-coordinated Ti surface atom [331]. Again, this is similar to the spin- resolved calculations for the defect state on the clean surface, Fig. 37.

Interestingly, adsorbed Na promotes the adsorption of CO2 and the dissociation of NO. Neither 2À molecule binds to a clean surface at room temperature. Adsorbed CO2 forms a carbonate (CO3 ) complex at the surface. Onishi et al. [205,304,305] were the ®rst to observe that a critical Na coverage is necessarybefore this adsorption takes place at room temperature. This critical coverage coincides with the formation of an ordered c(4 Â 2) superstructure. This interesting promotion effect has prompted some work on the adsorption geometryof Na. The results are partiallycon¯icting and are summarized in Fig. 42. The ®rst structural model was introduced byOnishi et al. [205,304,305]. Based on XPS intensities, a Na coverage of 0.5 ML was calculated, and from the extinction of spots in LEED, a ``Na2O-dimer'' model was proposed (Fig. 42a). More recently, the same group has conducted STM experiments, and found bright spots on bright rows, i.e., on the position of the ®vefold coordinated Ti atoms (Fig. 42 [323]). Consequently,their former model was revoked bythe authors. A previous STM studywas performed on the TiO2(1 1 0)-(1 Â 2) surface [160]. Based on photoemission experiments, Nerlov et al. [324±326] suggested two possible adsorption sites for Na in tetrahedral coordination with three oxygen atoms, see Fig. 42c. A Na atom on the ``between'' site is bonded to two bridging and one in-plane oxygen atom, a Na atom in the ``adjacent'' site is bonded to one bridging oxygen and two in-plane oxygen atoms. Hartree± Fock and Monte Carlo simulations of Na shows a preference for ``between'' sites, see Fig. 42c. In ion shadowing/blocking measurements it is seen that adsorption of Na removes a peak that is due to scattering from the ®vefold coordinated Ti atoms [327]. Under the assumption of a `derelaxation' of the bridging oxygen atoms, the results are consistent with ``adjacent'' Na, but inconsistent with ``between'' Na. This is in disagreement with the theoretical results (Fig. 42d) published in [329,330]. A molecular dynamics simulation of Na atoms at different coverages showed preferred adsorption sites neighboring two bridging oxygen atoms, in `between' positions. However, a more recent calculation by the same group indicates that the position might depend on the Na coverage [319].AnadsorptionontheTisitesasproposedin[323] was not explicitlyconsidered [327], but such a position does not seen to be likelyon chemical grounds.

The proposed geometries of the carbonate ion that forms upon adsorption of CO2 on a Na-covered surface are also summarized in Fig. 42e and f. A linear con®guration was proposed based on STM 2À images, see Fig. 42e. The third oxygen in the CO3 complex is a bridging oxygen atom. NEXAFS 2À experiments [328] show that the molecular plane of CO3 is twisted by32 Æ 5 out of the [0 0 1] direction with a tilt of 46 Æ 5 awayfrom the surface normal, in disagreement with this model. Two different con®gurations have been suggested (Fig. 42f), one where CO2 reacts with the bridging oxygen atom (left side in Fig. 42f), and one where it occupies a site between the bridging O atoms (right side in Fig. 42f). In this case carbonate formation must involve abstraction of O from the selvedge [328].

4.2.3. Potassium Adsorption of K has been studied on rutile (1 0 0) and (1 1 0). The experimental results on the electronic structure are similar on both surfaces [333±335], and consistent with other alkali metals. K 126 U. Diebold / Surface Science Reports 48 (2003) 53±229 adsorption on the annealed TiO2(1 1 0) surface induces a Ti3d state in the band gap and a change in work function [333,334]. Theoretical results indicate that this state is spin-polarized [278,336]. IPS data show two distinct features at 1.7 and 5.3 eV above EF, associated with Ti3d-derived t2g and eg states, respectively [337] (see also Fig. 33). The electronic structure has also been probed with D ion scattering [322]. On the slightlydefective surface, a decrease in the initial concentration of O vacancy was observed and assigned to O diffusion to the surface, induced byK adsorption [334]. This is also consistent with the observation of K2O multilayers that grow by extracting oxygen from the substrate [335]. Heating a K-covered surface to a temperature above 1000 K causes a drastic reduction of the

TiO2(1 1 0) substrate [333]. No long-range ordering was observed on TiO2(1 1 0), making a structural assignment dif®cult. Calculations indicate a strong repulsive interaction between adsorbed K, and a preference for K to adsorb in the vicinityof the bridging oxygenatoms [336] in a `between' site (Fig. 42) for higher coverages. In contrast to the absence of a LEED pattern on TiO2(1 1 0), a TiO2(1 0 0)-c(2 Â 2)K overlayer forms at 0.5 ML coverage [338±340]. This layer desorbs at 750 8C [340]. SEXAFS measurements [338,339] indicate that K occupies a bridge site, probablybetween two of the twofold coordinated oxygens protruding from the TiO2(1 0 0)-(1 Â 1) surface (see Fig. 23A). This adsorption geometrywas con®rmed in a recent ab initio study [278]. Large relaxations of the neighboring O atoms were found.

The co-adsorption of O2 on TiO2(1 0 0) indicates the formation of a peroxide or superoxide at the surface [341]. The uptake of CO2 on the reconstructed TiO2(1 0 0)-(1 Â 3) surface, promoted with K and Cs, was monitored with MIES and UPS [342]. Similar as in the case of Na, the surface needs to be alkalated before anyuptake of CO 2 occurs at room temperature, and the formation of a CO3 has been invoked.

4.2.4. Cesium

The adsorption of Cs on a TiO2(1 1 0) rutile surface was investigated byGrant and Campbell [343]. The formation of a strongly bound, ionized Cs layer, followed by multilayers, was observed. From TPD and XPS data it was concluded that Cs displays a Stranski±Krastanov growth mode below room temperature, with the completion of a uniform ML, followed bythe growth of 3D clusters of Cs that cover onlya small fraction of the surface. The Cs in the ®rst 1/2 ML is verycationic, donating electron densityto the substrate. Most of this charge is localized near the topmost atomic layers,with Ti4 ions being reduced to Ti3. A local dipole moment of the adsorbate±substrate complex of 6 Debye at 0.1 ML was extracted from band-bending corrected work function measurements. A downward band bending of 0.2±0.3 eV occurs and saturates at 0.05 ML. A rapid and smooth decrease in the heat of adsorption with coverage from initially>208 kJ/mol (which is too high to be probed with TPD) down to 78 kJ/mol for 1 ML was observed. No ordered LEED structures were observed. Ã Metastable (He 1s2s) impact electron spectroscopy(MIES) and UPS results on Cs/TiO 2(1 1 0), deposited at room temperature, were reported byBrause et al. [344]. With increasing coverage, a Ti3d and a Cs6s feature in the band gap were identi®ed with UPS and MIES, respectively. The MIES results for low Cs coverages are compatible with fullyionic adsorption of Cs. For larger coverages the charge transfer is incomplete. The ionicity,i.e., the charge transfer per adsorbed Cs atom, decreases from practicallyunityto 50% and 12% at coverages of 0.5 ML and 1 ML, respectively. This is in agreement with earlier experiments [321,343]. After turning off the Cs supply, the MIES spectrum of a surface covered with Cs U. Diebold / Surface Science Reports 48 (2003) 53±229 127 changes slowly. This was interpreted as a rearrangement of the oxygen in the surface layer whereby the

Ti and Cs ions compete for the bonding to the oxygen. Cs promotes the adsorption of CO2 on a TiO2(1 0 0)-(1 Â 3) surface [342].

4.2.5. Calcium

The interaction between Ca ions and TiO2 plays a major role for the performance of Ti-containing steels as bone implants. No studyof the deposition of Ca on TiO 2 has been performed. However, Ca impurities in the bulk of the crystal tend to segregate towards the surface upon annealing [145±148].The interactionofsuchsegregatedCaimpuritiestothesurfaceofTiO2(1 1 0) is discussed in Section 2.2.1.4.

4.2.6. Aluminum The interactions of ultrathin Al ®lms with both stoichiometric and oxygen de®cient (sputtered)

TiO2(1 1 0) surfaces were studied with several electron spectroscopies and LEED byDake and Lad [345]. The interfacial reactions in aluminum ®lms resemble those of the earlytransition metals. It interacts stronglywith both stoichiometric and sub-stoichiometric TiO 2(1 1 0) surfaces at room temperature. For monolayer Al doses, the Al layer is oxidized, the Ti ions in the TiO2 substrate underneath reduced, and the long-range order of the substrate is lost. The interfacial oxidation/ reduction reaction occurs also on pre-reduced TiO2 surfaces. No evidence for an intermetallic Al±Ti alloyat the interface was found. An aluminum oxide ®lm continues to grow at higher doses by extracting O anions from the TiO2 substrate. When oxygen diffusion becomes rate-limiting, a heterogeneous mixture of aluminum oxide and metallic aluminum is created on top of the oxygen- de®cient TiO2Àx substrate. This mixed layer is unstable upon heating and converts completely into aluminum oxide. In addition, the annealing treatment causes a reoxidation of the reduced substrate. STM results of the initial stages of the growth are consistent with the growth mode deduced from spectroscopic results. A disordering of the substrate is seen for verysmall amounts of deposited Al [346,347]. Small clusters were visible on the terraces but no preferred nucleation sites were observed. STM studies of Al deposited on a faceted rutile (1 0 0) surface showed island growth [348].

The adsorption of Al on chemicallymodi®ed TiO 2(1 1 0) surfaces has also been studied [281,349]. Aluminum deposited onto a TiO2 surface pre-dosed with 1 ML potassium competes with the potassium for oxygen, and reduces additional Ti4 cations in the interface region, with the possible formation of a Ti±O±Al±K complex. Carbon species adsorbed on the TiO2(1 1 0) surface byelectron beam decomposition of C2H4 prior to the Al deposition interact onlyweaklywith the oxide surface, but retard the oxidation rate of Al and decrease the temperature stabilityof the Al 2O3 overlayer [349].

4.2.7. Titanium

Rocker and GoÈpel [350] showed that non-stoichiometric Ti-rich TiO2 surfaces can be prepared byTi evaporation. Spectroscopically, the resulting ®lms resemble a surface reduced by sputtering. The growth of Ti was investigated byMayeret al. [280] with XPS and LEIS. Ti interacts with the TiO2 substrate, and forms a layer of reduced TiOx that is thicker than the deposited ®lm. From LEIS data it was concluded that Ti clusters form on top of this ®lm. The thermal stabilityof TiO x overlayers has been investigated byHenderson [74]. From SSIMS experiments with isotopicallylabeled 46Ti and 18O it was concluded that diffusion at the surface takes place between 400 and 700 K, and that diffusion into the bulk starts at temperatures >700 K. The predominant diffusing species was identi®ed as Ti cations, not O anions/vacancies. 128 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 43. (a) Soft X-rayphotoemission spectra of Hf overlayerson TiO 2(1 1 0) (photon energyof 110 eV). The spectra are deconvoluted to show the contributions of different oxidation states. The intensities of the contributions to the total peak area are shown in (b). The initial growth mode is depicted schematicallyin (c). From [351].

4.2.8. Hafnium Hafnium is a veryreactive material, and the growth mode as determined with soft X-ray photoemission and supporting LEIS measurements is summarized in Fig. 43 [351]. Upon Hf deposition at room temperature, an oxidized layer of HfO2 is formed on top of a reduced layer of TiO2Àx. LEIS data clearlyindicate that this layercovers the surface completely [63], and that oxygen migrates from the substrate into the overlayer. The substrate LEED pattern disappears very quickly, and no long-range ordered structure is formed. For thick layers, metallic Hf is observed, probably in the form of metallic clusters. This metallic Hf is oxidized upon heating [63]. A schematic diagram of the initial stages of the growth is drawn in Fig. 43.

4.2.9. Vanadium

Room-temperature deposition of metallic V shows a vigorous reaction with the TiO2 substrate. A clear Ti3 feature evolves, and the LEED pattern of the clean surface disappears for small coverages [14,15,352±354]. Concurrent with the reduction of the substrate, the V overlayer is oxidized, probably in a V3 oxidation state [352]. For thicker ®lms, metallic V forms. Metallic V was also observed when vanadium was deposited onto a sputter-reduced surface [15]. Electron diffraction experiments were performed on a ®lm with a coverage of a fractional monolayer after annealing at 473 K. The results are consistent with an adsorption site underneath the bridging oxygen atoms, at the position of the sixfold coordinated Ti atoms [14]. At higher coverages (5 ML) V islands form with a bcc (1 0 0) structure.

No long-range ordering is visible, but epitaxywith alignment bcc(1 0 0)[1 0 0]V||(1 1 0)[1 1 0]TiO 2 is U. Diebold / Surface Science Reports 48 (2003) 53±229 129

Fig. 44. STM images (250 Ð Â 250 Ð, 2.0 V, 0.1 nA) of a clean (a) and vanadium-covered TiO2(1 1 0)-(1 Â 2) surface: (b) 0.05 ML V, (c) 0.2 ML V, (d) 0.6 ML V. From Biener et al. [12]. # 1999 Elsevier. concluded from ARXPS measurements. This is similar to other bcc metals on TiO2(1 1 0) [355]. When oxygen is dosed on metallic V/TiO2(1 1 0) surfaces, the ®lms are oxidized, probablyto V 2O3 [352]. Vanadium ®lms, evaporated on a TiO2(1 1 0)-(1 Â 2) surface at 300 K were studied bythe group of Madix [12,354,356,357]. The V binds preferentiallyon top of the (1 Â 2) strands with formation of small clusters, see Fig. 44. This maximizes the contact with surface oxygen (if one assumes an added

Ti2O3 row model for the (1 Â 2) reconstruction, see Fig. 18b). No preferential adsorption at step edges is visible. As the vanadium coverage is increased toward one monolayer, the density of vanadium clusters decorating the (1 Â 2) rows increases, forming chains of more or less isolated clusters. At three monolayers coverage vanadium metal forms a granular ®lm which covers the oxide surface uniformly. The ®lm is thermallynot stable with oxidation occurring at 500 K and the onset of diffusion into the bulk around 600 K [12,356]. Similar results have also been obtained on a TiO2(1 1 0) surface [354]. Atomicallyresolved images of subsurface impurityatoms have been attributed to substitutional V atoms [153].

4.2.10. Vanadia Deposition of vanadium in an oxygen ambient results in vanadium oxide that interacts only weakly with the support [15,272]. The Ti2p levels remain fullyoxidized. The work function is increased [15,272]. 130 U. Diebold / Surface Science Reports 48 (2003) 53±229

The stoichiometry and short-range order of vanadia overlayers can be controlled by varying the deposition parameters. When V is deposited in an oxygen pressure of 2 Â 10À6 Torr at room temperature, XPS and valence band photoemission as well as SEXAFS results are compatible with a

V2O3 overlayer [272]. No long-range ordering is visible with LEED. The oxidized overlayer is much more stable than a metallic V ®lm; annealing at 1100 K causes a structural re-arrangement with formation of larger vanadia particles and formation of a VTiO3 layer at the interface [272],butno diffusion into the substrate.

AVO2 layer can be grown by successive cycles of fractional-ML vanadium metal deposition À6 followed byannealing at 473 K in 2 Â 10 mbar O2 [358]. The V3p photoemission peak consists of two distinct components chemicallyshifted by1.3 eV. In angle resolved photoemission extended ®ne structure (ARPEFS) scans, the lower BE signal, associated with the VO2 phase, shows well de®ned intensitymodulations whose main features are similar to the ARPEFS scan of the Ti3p signal of the substrate. This indicates formation of an ordered VO2 phase with a rutile structure epitaxial to the substrate. In addition a highlyoxidized, less ordered phase forms.

The reaction to VO2 can be stopped at VOx (x 1) when annealing of the metal ®lm is carried out in UHV [359]. Under speci®c conditions, locallyordered VO x ®lms with a thickness of up to 5 ML can be prepared. A detailed structural analysis was given by Negra et al. [360]. A NaCl-like stacking of the overlayer with (1 0 0) orientation and alignment of the [0 0 1] overlayer azimuth with the substrate 112 direction was concluded from X-rayphotoelectron measurements.

Methanol is oxidized to formaldehyde on a TiO2(1 1 0)-supported vanadium oxide monolayer (probablyV 2O3), whereas both clean TiO2(1 1 0)- and TiO2(1 1 0)-supported vanadia multilayers are inactive for the reaction [361,362]. This result is similar to the reactivitytrends found for high-surface area catalysts.

4.2.11. Niobium The growth of Nb overlayers was investigated by Marien et al. [363±365]. When Nb ®lms are grown at room temperature, a reacted interface forms as is expected for an earlytransition metal. Direct evidence for this structurallydistorted interlayeris given bythe high-resolution transmission electron microscopy(HRTEM) image in Fig. 45. The about 2 nm thick interlayer contains many structural defects, but the image contrast clearlyshows that it did not transform into an amorphous state. Electron- energyloss spectra have been taken with a scanning transmission electron microscope (STEM) in steps of 0.5 nm across the interface (Fig. 45). The shape of the TiL23 edge is verysensitive to changes in oxidation states and local bonding geometries, and the evolution of the spectra across the interface is consistent with a TiO2Àx layer depleted of oxygen. The ®rst two monolayers of Nb adjacent to the substrate are oxidized. The interdiffusion of Nb with the TiO2 substrate is kineticallyhindered at the low deposition temperature used. On top of the interfacial reaction layer, Nb grows epitaxially with its

(1 0 0) planes parallel to TiO2(1 1 0) and the [1 0 0] overlayer azimuth aligned with the substrate [0 0 1] direction.

The epitaxial growth of Nb-doped TiO2 ®lms on (1 0 0) and (1 1 0)-oriented substrates has been discussed in a series of papers from Chambers' group [93,190,366±374]. Nb substitutionally incorporates at cation sites in the rutile lattice, forming NbxTi1ÀxO2 solid solutions. Analysis of the Nb 4 3d5/2 core-level binding energysuggests a Nb oxidation state. The crystal quality and surface roughness of the ®lms depend stronglyon the substrate orientation. The epitaxial ®lms grow in a layer- by-layer fashion and have excellent short- and long-range structural order at x 0:3onTiO2(1 1 0) and U. Diebold / Surface Science Reports 48 (2003) 53±229 131

Fig. 45. (a) HRTEM micrograph of the Nb/TiO2(1 1 0) interface. A distorted interlayer between the Nb and TiO2 crystal is present. Due to the high defect concentration in this region, the atomic structure of this interlayer is poorly resolved. (b) Line pro®le of EEL spectra measured perpendicular to the interface bySTEM. The spectra show a sharp transition from the interlayer of the perfect rutile (top curve) with crystal-®eld split Ti-L23 edges. The step width of the electron beam was 0.5 nm. From Marien et al. [365]. # 2000 Elsevier. 132 U. Diebold / Surface Science Reports 48 (2003) 53±229 at x 0:1onTiO2(1 0 0). At higher doping levels dislocations form and the ®lms become rough. The Nb±O bond lengths in NbxTi1ÀxO2 are the same as the Ti±O bond lengths in pure TiO2 prior to the onset of dislocation formation. The Nb substitution for Ti in the lattice introduces an additional valence electron per atom. These extra Nb electrons form a non-bonding band which is degenerate with the valence band; no new state densitywas found experimentallyin either the band gap or conduction band. The effect of Nb doping on the band gap was studied theoretically [58,375]. In these investigations

Nb is found to introduce a shallow donor level in bulk TiO2 [375]. An impurityrelated state is predicted to occur in the gap of Nb-doped TiO2(1 1 0) surfaces, in contradiction to the photoemission experiments quoted above [93,190,366±374].A5 rather than a 4 oxidation state for surface cations was predicted. The structural changes undergone bythe (1 1 0) surface upon doping are small and insensitive to the impuritycharge state. An experimental studybyMorris et al. [376] showed core-level spectra consistent with a Nb5 oxidation state and a gap state with Ti3. STM images of Nb-doped

TiO2(1 1 0) showed bright spots that were attributed to a delocalization of charge to the neighboring atoms [376]. The dispute about the charge state of Nb in TiO2 is not resolved at this point. Possiblyit depends on the sample growth conditions.

4.2.12. Chromium The growth, interfacial reaction, ®lm structure and thermal stabilitywas investigated in a series of papers byPan et al. [292,355,377±380]. The overall growth behavior ®ts into the categoryof early transition metals, with an intermediate layer formed upon chromium deposition, and `wetting' of the substrate byan oxidized chromium overlayer. The wetting correlates with the formation of reduced Ti species as observed with XPS. Low-energyion scattering experiments with isotopicallylabeled 18O showed that this oxidation/reduction reaction is due to a dynamic incorporation of lattice oxygen into the chromium overlayer. The same epitaxial relationship as already discussed for Nb and V, i.e., bcc(1 0 0)[1 0 0]Cr||(1 1 0)[1 1 0]TiO2 was also observed for chromium overlayers. Chromium is soluble in TiO2 and chromium-doped TiO2 ceramics are used in varistors. The annealing behavior of vapor- deposited ®lms is complex, with clustering of metallic chromium competing with diffusion into the bulk. Prolonged annealing at high temperature of chromium overlayer leads to dissolution into the bulk. The band gap width and the charge state of chromium dopants was calculated bySambrano et al. [58].

TiO2(1 1 0) is a useful substrate for the growth of CrO2, which also has the rutile structure. CrO2 has properties that might be of interest for spintronics applications, as it is expected to be a half-metallic ferromagnet [381].

4.2.13. Molybdenum

The deposition of metallic molybdenum on TiO2(1 1 0) has been studied with AES, XPS, ex situ AFM, and RHEED [382]. The in¯uence of different substrate preparations were tested. A stoichiometric surface was produced bysputtering and annealing in oxygen,a non-stoichiometric, rough surface byAr sputtering without annealing, and rough but stoichiometric surface bysputtering with O. In each case, in situ AES and XPS studies and ex situ AFM and RHEED characterizations indicate a Stranski±Krastanov growth. After completion of three monolayers, island growth is observed. The ®rst three monolayers consist of amorphous Mo oxide with a Mo oxidation state between Mo3 and Mo4. The oxidation of the Mo layers generates a reduction of the substrate with the formation of Ti3 and Ti2 and induces a reconstruction of the surface. During the formation of the Mo oxide layers, the roughness of the surface strongly decreases. After the growth of the three layers, the U. Diebold / Surface Science Reports 48 (2003) 53±229 133 surface is ¯at whatever the initial roughness. The resulting islands are metallic (bcc structure) but without a preferential orientation.

4.2.14. Molybdena The preparation of ®nelydispersed Mo oxide species, and their characterization with polarization dependent total re¯ection ¯uorescence EXAFS has been discussed in several papers [383±390].A model sample for supported Mo oxide catalysts was prepared by impregnation of (NH4)6Mo7O24Á4H2O dissolved in ultrapure water, followed bycalcination at 773 K. The structure and orientation of Mo species are stronglyaffected bythe pretreatment conditions and impurities. Mo dimers oriented to [1 1 0] direction or a Mo chain structure along [0 0 1] are present when theyare prepared under oxidative conditions or reductive conditions, respectively. Mo tetrahedral monomers are formed when Na and K are present on the surface.

The deposition of molybdenum oxide monolayers by heating a TiO2(1 1 0) crystal surface covered by MoO3 powder was also explored and investigated with XPS and electrochemical methods [391]. Deposited Mo ®lms are progressivelyoxidized when heated at several hundred degrees centigrade, with the stoichiometrydepending on the annealing conditions [392].

4.2.15. Manganese X-rayabsorption and photoelectron spectroscopies were used to studythe adsorption and reaction of

Mn ®lms deposited on TiO2(110)at258C and after annealing to ca. 650 8C [271].Fractionalmonolayer coverages of Mn at 25 8C produce a reactive, disordered interface consisting of reduced Ti cations and oxidized Mn overlayer atoms. The electronic structure of the ®lm resembles MnO, and the Mn L-edge adsorption spectra clearlyindicate a Mn 2 oxidation state. Metallic Mn is found onlyfor thicker layers. Annealing Mn ®lms to ca. 650 8C leads to several changes that are largelyindependent of initial overlayer coverage: metallic Mn thermallydesorbs leaving onlyMn 2 ions; interfacial Ti cations are largelyre- oxidized to the Ti4 state; and the local order is increased at the interface. The formation of a crystalline ternarysurface oxide, MnTiO x, was proposed to account for these chemical and structural changes [271].

4.2.16. Manganese oxide

Oxygen plasma assisted MBE was used to grow epitaxial ®lms of pyrolusite (b-MnO2, rutile structure) on TiO2(1 1 0) for thicknesses of one to six bilayers (BL) [393]. In this work, one bilayer was de®ned as a layer of Mn and lattice O and an adjacent layer of bridging O within the rutile structure, see Fig. 6. The resulting surfaces have been characterized in situ byRHEED, LEED, XPS, XPD, and AFM.

Despite a lattice mismatch with rutile TiO2 of 3±4%, well-ordered, pseudomorphic overlayers form for substrate temperatures between 400 and 500 8C. Mn±Ti intermixing occurs over the time scale of ®lm growth (ca. 1 BL/min) for substrate temperatures in excess of 500 8C. Films grown at 400±500 8C exhibit island growth, whereas intermixed ®lms grown at temperatures of 500±600 8C are more laminar. One BL thick ®lms grown at 450 8C are more laminar than multilayer ®lms grown at the same temperature, and form a well-ordered surface cation layer of Mn on the rutile structure with at most 10% indiffusion to the second cation layer.

4.2.17. Iron

The ®rst growth studyof iron on a TiO 2(0 0 1) surface was performed byBrugniau et al. [394]. During growth of the ®rst layer, a decrease in the work function, changes in secondary electron 134 U. Diebold / Surface Science Reports 48 (2003) 53±229 emission, and changes in Auger features were observed. On TiO2(1 1 0) a layer-by-layer growth mode was reported byDeng et al. [395]. CO was weaklyadsorbed on a low Fe coverage system(0.2 ML) and adsorbed stronglyto thicker Fe layers. Low-energyion scattering measurements from Madeyand co-workers [379] showed a cluster growth when iron was deposited at room temperature, although Fe tends to wet the surface better than late- transition metal overlayers, see Fig. 41. A (not verystrong) oxidation/reduction reaction occurs at the interface. Complete coverage of the substrate can be reached when a 1 ML ®lm is oxidized bydosing oxygen. UPS measurements indicate a mixed iron-oxide layer composed of FeO and Fe2O3 phases [396]. The surface roughness also plays a role in the morphology of the metallic ®lms [379]. Mostefa- Sba et al. [397] observed a 2D growth mode up to three monolayers for a high initial roughness of the substrate. The orientation of the metal overlayer was determined as bcc(1 0 0)-oriented ®lms with diffraction techniques sensitive to both long-range and short-range order [355]. As is the case for other bcc overlayers (Cr, Nb, and V) the overlayer [0 0 1] direction is aligned with the substrate [0 0 1] direction.

The electronic structure of Fe/TiO2(1 1 0) was studied with inverse photoemission, UPS [396], and resonant photoemission [262]. As seen in Fig. 46, band gap states are induced byfractional Fe coverages on stoichiometric TiO2(1 1 0) surfaces. The higher lying band gap feature shows the same

Fig. 46. Left panel: photoemission from the valence band of (a) stoichiometric TiO2(1 1 0) and (b)±(f) after deposition of Fe at room temperature onto a stoichiometric surface. Fe coverages (in equivalent ML) are (b) 0.16 ML, (c) 0.24 ML, (d) 0.4 ML, and (e) 1.3 ML. Spectrum (f) was taken after irradiation of an uncovered, stoichiometric surface with 3 keV electrons. All spectra are taken with a photon energyof 110 eV. Right panel: intensityof Fe3d-derived (circles) and Ti3d-derived (triangles) valence band features from 0.2 equivalent ML Fe on stoichiometric TiO2(1 1 0). The nature of the band gap states can be determined from their behavior in these resonant photoemission curves. From [262]. U. Diebold / Surface Science Reports 48 (2003) 53±229 135 resonance behavior as an oxygen-vacancy induced Ti3d band gap state. The resonant line shape of the feature at higher energyis verysimilar to the shape observed for iron oxides [398,399]. After thermal annealing in UHV at 500±700 8C for several hours, Fe clusters are almost completely covered byTi suboxides TiO x while Fe remains mainlyin the metallic state. The Fe signal disappears in LEIS, and re-emerges after small ion ¯uences during sputtering [17]. As mentioned above, this encapsulation process (SMSI) is common to all group VIII A metals studied (Fe, Co, Rh, Ir, Ni, Pd, Pt).

4.2.18. Ruthenium

The thermal decomposition of Ru3(CO)12 has been used as a source for the growth of metallic ruthenium particles and RuO2 ®lms on a TiO2(1 1 0) surface [400±402]. After adsorption at room temperature and heating in UHV, the precursor completelydecomposes at 300 8C. The deposited metal particles show some residual carbon contamination and are disordered, both in the short and long range. The high XPS binding energyfound for the metal core levels suggests the presence of nano-clusters on the surface. The titanium atoms of the substrate are not reduced byreaction with the incoming metal atoms. Thermal treatment of the surface at 400 8C forms a Ti1ÀxRuxO2 surface compound [401]. The `reverse' system, Ti and TiO2 ®lms on RuO2(1 1 0) and (1 0 0) was investigated with XPS, AES, and LEED [403]. RuO2 also has rutile structure, and epitaxial TiO2 layers could be achieved at 600 8C. Ru±RuOx-promoted TiO2 was shown to be a good catalyst for the selective methanation of CO2 at room temperature and atmospheric pressures [404].

4.2.19. Ruthenium oxide

RuO2 is a metallic oxide with the rutile structure and a lattice mismatch with respect to TiO2 of 4.7 and À2.2% along the [0 0 1] and 1 10 directions, respectively. Epitaxial ®lms on TiO2(1 1 0) were grown with oxygen-plasma assisted molecular beam epitaxy (OPA-MBE) in Chambers' group [289].

Decomposition of Ru3(CO)12, while heating in an oxygen atmosphere, also results in epitaxial RuO2 ®lms [400,402]. As mentioned above, the RuO2/TiO2 interface is thermodynamically not stable and prone to intermixing. This has been exploited for growing a `graded' interface, essentiallya crystalline stack of the form RuO2=RuxTi1ÀxO2=TiO2 110 where x increased with thickness. In this way, a smooth ®lm was achieved [289].

4.2.20. Cobalt

Metallic cobalt and cobalt oxide deposited on oxidized and sputtered TiO2(1 1 0) was investigated by Shao et al. [405]. The Ti2p spectra show a small reduction upon Co deposition at room temperature, and a pronounced Ti3 peak after heating at 900 K. The latter is consistent with other group VIII metals that show the SMSI effect. The doping of TiO2 (anatase) ®lms with Co has received some interest recently because such ®lms are conductive, ferromagnetic, and opticallytransparent at room temperature [55]; these properties make it a possible candidate for spintronics applications [56]. In the weaklyferromagnetic ®lms the Co is bound in the lattice in a substitutional form and exhibits in a 2 oxidation state. The interaction of aqueous Co(II) species with the (1 1 0) and (1 0 0) surface of rutile has been investigated with grazing-incidence XAFS spectroscopyunder ambient conditions in a humid atmosphere [406]. The Co(II) surface complexes adsorbs at sites corresponding to Ti-equivalent positions in an extension of the rutile structure. This result suggests that even if different crystallographic surfaces of metal oxides have strongly differing adsorption properties for gaseous species in ultra-high vacuum, theycan have similar properties for adsorption of metal ions in aqueous 136 U. Diebold / Surface Science Reports 48 (2003) 53±229 solution. No evidence was found for well-ordered Co(II)-hydroxide-like precipitates that would show Co±Co pair correlations, or for Co±Ti pair correlations.

4.2.21. Rhodium

TiO2(1 1 0). A series of growth studies on TiO2(1 1 0) surfaces was published byBerkoÂ et al. [407± 409]. Rhodium was deposited on a TiO2(1 1 0)-(1 Â 2) surface at room temperature and investigated with AES and STM. Nucleation of Rh particles occurs at terraces (no step decoration) [407].Atlow coverages (0.01 ML), small clusters with ca. 2±6 atoms/cluster were observed, with a preferential location at the bright rows. A slight increase in cluster size was experienced after several hours, even at room temperature. Rhodium deposited on an oxidized Ti foil did not show anyevidence of an interfacial oxidation/reduction reaction in XPS [410].

One of the motivations for surface-science studies of the Rh/TiO2(1 1 0) system was to investigate the SMSI state of rhodium-based catalysts. BerkoÂ et al. [408] found in an STM/XPS studythat the surface pretreatment in¯uences whether or not encapsulation occurs. A stoichiometric and well-ordered

TiO2(1 1 0)-(1 Â 1) surface covered by3 ML of rhodium showed no evidence for encapsulation in the temperature range of 300±800 K in UHV [408]. A few minutes annealing in a H2 atmosphere of 10À4 mbar at 750 K did produce encapsulation with a reduced overlayer. Pretreatment of the stoichiometric TiO2(1 1 0)-(1 Â 1) surface byAr bombardment (creation of surface and subsurface Ti3 states) before Rh deposition resulted in encapsulation of Rh particles after annealing in UHV (in the absence of hydrogen). It was concluded that the presence of lattice defects in the near-surface region is crucial for the SMSI state to form. On oxidized Ti ®lms, monolayer coverages of Rh encapsulate upon UHV annealing around 750 K. Above 820 K, Rh diffuses into the substrate [410]. The encapsulation process between 500 and 700 K results in an increase of the size of clusters at very low Rh coverages (0.01 ML). At higher Rh coverages annealing does not cause appreciable changes in STM images. Annealing >1100 K causes the separation of the 3±5 nm wide and 3±5 atomic-layers thick Rh crystallites (with their (1 1 1) plane parallel to the substrate), probably due to the de- encapsulation of the Rh crystallites [407]. This thermal behavior was exploited to independentlycontrol the particle size and distance of Rh clusters on TiO2(1 1 0)-(1 Â 2) [409]. In a `seeding' step, a small amount of Rh (0.001±0.050 ML) was deposited at room temperature and annealed at 1100 K. Further Rh deposition caused growth of the Rh nanoparticles and the formation of a narrow size distribution. The mechanism of this procedure is based on the large difference in the surface diffusion coef®cient between Rh adatoms and Rh nanocrystallites larger than 1±2 nm. In the seeding step the average distance between the metal particles is controlled, the second step determines the particle size (2±50 nm). Growth at 1100 K resulted in crystallites with a well-de®ned shape, either hexagonal or elongated in the [0 0 1] direction.

When these large Rh crystallites (diameter of 10±15 nm) were annealed in H2 at 750 K, dramatic morphological changes (corrosion or disruption) were observed to accompanythe encapsulation process [407].

Highlydispersed rhodium clusters on TiO 2(1 1 0) were grown with MOCVD [411,412]. A rhodium gem-dicarbonyl species, Rh(CO)2, was prepared bythe dissociative adsorption of {Rh(CO) 2Cl}2. Desorption of CO at 500 K, or reaction with hydrogen at 300 K, results in the formation of a highly 0 dispersed Rhx species from which the gem-dicarbonyl can be partially re-generated by exposure to CO. Heating the rhodium overlayer to higher temperatures leads to the nucleation of larger metallic particles from which the gem-dicarbonyl cannot be re-formed by exposure to CO. U. Diebold / Surface Science Reports 48 (2003) 53±229 137

TiO2(0 0 1). Growth of Rh on TiO2(0 0 1) at room temperature also showed 3D particles [413]. The particles showed a narrow size distribution, and coalesced upon annealing.

4.2.22. Iridium

The growth of Ir on TiO2(1 1 0)-(1 Â 2) was studied byBerkoÂ et al. [414,415]. At verylow coverage Ir forms round nanoparticles which are mainlycentered on the rows of the (1 Â 2) terraces. Annealing of the iridium covered surface caused an perceivable increase of the particle size onlyabove 700 K. The supported iridium nanoparticles of 1±3 nm exhibit a veryhigh reactivitytowards CO. As a result of the CO adsorption at 300 K, crystallites disrupt into smaller particles, and ®nally into atomically dispersed

Ir. This feature was not observed for larger clusters of 8±10 nm size. Similar to the Rh/TiO2(1 1 0)- (1 Â 2) system, a method was developed to synthesize Ir nanoparticles in desired uniform sizes in the range 1.5±20.0 nm with constant interparticle distances [414]. The method consists of two steps: (i) vapor deposition of Ir metal in predetermined concentrations on titania at 300 K with a post-deposition annealing at 1200 K, and (ii) subsequent evaporation of Ir on this surface at 1200 K.

4.2.23. Nickel

TiO2(1 1 0). The growth of nickel on TiO2(1 1 0) was investigated with spectroscopic methods [416±421], and, more recently, with STM [422] and computational approaches [423]. Based on AES measurements, a S±K growth mode was postulated [417,419]. A ®rst-principles study of the initial stages of Ni growth [423] indicated that Ni adsorbs preferentiallyon top of bridging oxygen atoms and that the bond strength between Ni adatoms and the substrate is much stronger than between Ni adatoms. EXAFS experiments on Ni/TiO2(1 1 0) also indicated onlytwo-dimensional (2D) growth of Ni [421]. A model was developed with Ni atoms forming chains in the channels determined byO atoms. However, more recent STM images clearlyindicate a cluster growth at low coverages, in line with thermodynamic expectations [422]. At low coverages, the clusters appear round-shaped in STM, and substrate step edges act as nucleation sites at verylow coverages. At higher coverages, the clusters expose well-de®ned facets. Differing results on overlayer geometry are reported as well. XPD measurements indicated both, (1 0 0)- and (1 1 1)-oriented islands [417], and a LEED studywas interpreted as hexagonal, (1 1 1) oriented islands, both parallel to TiO2(1 1 0) and tilted [418]. `Hut-clusters' with the base consisting of (1 1 0)-oriented layers have been observed with STM and RHEED. The long and short sides of the `huts' consist of {1 1 1}- and {1 0 0}-oriented facets [422]. Electron transfer from Ni to the substrate in the order of ca. 0.1eÀ/Ni atom was estimated from spectroscopic measurements [416,417]. Electron donation of 0.37 and 0.27eÀ/Ni atom to the bridging and in-plane oxygen atoms was calculated by Cao et al. [423]. At fractional monolayer coverages the charge transfer causes a decrease in work function, band bending, and a shift in XPS and AES levels. A decrease in the XPS signal upon annealing was interpreted as diffusion of Ni into the substrate

[416]. STM shows coalescence and coarsening of Ni clusters on TiO2(1 1 0) up to a temperature of 880 K, the morphologyis stable above this temperature [422].

The adsorption of CO on Ni/TiO2(1 1 0) ®lms was studied byOnishi et al. [417] and in Mùller and Wu [419]. CO adsorbs molecularlyat all coverages. When a Ni ®lm on TiO 2(1 1 0) was exposed to air, it transformed into an epitaxial NiO overlayer. Reduction in UHV re-gained crystallographic spacings indicative of metallic Ni, but a NiO layer at the interface was postulated to account for the orientation of these clusters [422]. 138 U. Diebold / Surface Science Reports 48 (2003) 53±229

TiO2(1 0 0). Nickel deposition on TiO2(1 0 0) was studied byBourgeois et al. [424±426]. The results were interpreted in terms of a S±K type growth mode with three nickel layers completed before clustering starts.

4.2.24. Palladium

TiO2(1 1 0). The growth of Pd/TiO2(1 1 0) was investigated with STM [162,297,346,427±429] and spectroscopic techniques [10]. The growth mode is undoubtedlyVW-like. When depositing at room temperature, the cluster size increases with coverage [162]. A marked tendencyfor nucleation of Pd clusters at step edges was observed, similar to the case of Ni growth. The nucleation and cluster site is in¯uenced bythe defect concentration, however, with more random distribution across the surface on a highlydefective TiO 2(1 1 0) surface [297]. From STM images at low coverages, nucleation at the ®vefold coordinated Ti site was deduced [162]. A quantum-chemical DFT studymodeled the Pd/

TiO2(1 1 0) system with Pd atoms and dimers adsorbed on TiO2 clusters embedded in point charges [430]. In this theoretical studythe preferred adsorption for Pd was on the bridging oxygenatoms for small coverages, and along the Ti rows for higher coverages. Pd forms a covalent bond, slightly polarized towards the surface, but without signi®cant charge transfer. In the calculations, Pd dimers were not stable due to the strong bond with the substrate [430]. However, features in STM images were interpreted as dimers and trimers [162].

The geometryof thin ®lms of Pd deposited on the TiO 2(1 1 0) surface were investigated using coaxial impact-collision ion scattering spectroscopy(CAICISS) and RHEED [431]. Palladium islands in the Ê range from 5 to 40 A grew epitaxiallyon the TiO 2 surface with the orientation relationship of fcc 1 21 111jj001 110TiO2. This orientation relationship was observed in the temperature range from room temperature to 1170 K. Clusters with predominantly(1 1 1) orientation would be in agreement with the general trend for noble metal growth on TiO2(1 1 0), however, other orientations have been suggested for verysmall coverages (see below). The thermal behavior of Pd was investigated byseveral groups. An STM studybyStone et al. [427] observed aggregation of the overlayer clusters, as well the formation of a (1 Â 2) overlayer on the substrate after annealing to 773 K. This is supported byan STM studyfrom Frenken and co-workers [429]. The decrease of the Pd intensityin AES was interpreted as due to sintering rather than encapsulation [427]. LEIS experiments in Persaud and Madey [10] showed encapsulation. A simultaneous appearance of a (2 Â 1) superstructure and the encapsulation of the Pd islands was found after annealing at 1170 K in UHV byion scattering techniques [431]. In the process of encapsulation, the crystal structure of the Pd islands remained unchanged. This suggests that the encapsulating material is transported bysurface migration on the Pd particles. Heating in an O 2 atmosphere of 5 Â 10À5 Torr did not remove the encapsulating material. Evidence for encapsulation upon annealing

Pd/TiO2(1 1 0) to 800 K in UHV was also found in an adsorption experiment [432]. Similarly con¯icting results were found for Pt, where some research groups reported encapsulation upon UHV annealing, and others did not observe this effect. Possibly, these con¯icting results could be resolved similarlyas in the case of Pt (see below), and the reduction state of the sample could be a decisive factor on whether or not encapsulation occurs. Heating as-deposited Pd clusters in oxygen has dramatic effects as observed in a high-temperature

STM studybyBennett et al. [19,172]. The Pd nanoparticles on sub-stoichiometric TiO2(1 1 0) dissociativelyadsorb O 2 at 673 K which ``spills over'' onto the support where further reaction takes place. The spillover oxygen re-oxidizes the surface by removing Tin interstitial ions trapped in the U. Diebold / Surface Science Reports 48 (2003) 53±229 139 crystal lattice, preferentially re-growing TiO2 around and over the particles. This process is similar to the surface restructuring in an oxygen atmosphere described in Section 2.2.2.2, and was modeled in [433]. The adsorption of CO, and the corresponding bands in FT-RAIRS, was used to identifythe orientation of the Pd cluster sides [434]. STM work in Goodman and co-workers [162] con®rmed that CO adsorption itself does not change the morphologyof the clusters, although a CO background pressure during deposition did change the nucleation behavior. In the vibrational studyof CO/Pd/

TiO2(1 1 0), at small coverages the Pd surfaces appeared to be constituted of mainly(1 0 0)- or (1 1 0)- type sites. When annealed to 500 K, the Pd clusters coalesce and CO adsorbs predominantly as a stronglybound linear species which was associated with edge sites on the Pd particles [434].At coverages above 10 ML, the palladium particles exhibit (1 1 1) facets parallel to the substrate and aligned with the TiO2(1 1 0) unit cell, as was also observed byother researchers. This ordering in the particles is enhanced byannealing. The sintering process of Pd clusters at high temperatures was also observed directlywith STM [162,428]. Theoretical results of CO/Pd2/TiO2 (cluster) indicate that polarization effects cause the CO molecule to be less stronglybound as compared to a free Pd 2 dimer [430].

Adsorption of formic acid on Pd-promoted TiO2(1 1 0) was studied with HREELS [432]. In general, group VIII metals dissociate formic acid, and observed bands of formate species were attributed to adsorption at the bare areas of the substrate (see Section 5.2.1). Additional C±O species correspond to CO bound to the Pd clusters, and to CO and HCO at the cluster perimeters.

TiO2(1 0 0). On the TiO2(1 0 0)-(1 Â 3) surface, STM clearlyrevealed VW-like cluster growth [435]. At a coverage of 0.01 ML-equivalent, 35 AÊ wide and 8 AÊ high clusters were observed. At higher coverage the clusters coalesce. Theydo not order within the troughs ( Fig. 23) of the (1 Â 3)- reconstructed surface.

4.2.25. Platinum

Platinum is one of the most studied metal overlayers on single-crystalline TiO2. The interest has been stimulated bythe promotion effect of Pt in photocatalysis( Section 5.3), its use as an oxygen gas-sensor system, and by the fact that it is the classic SMSI system.

TiO2(1 1 0). As other late-transition metals, platinum does not cause an oxidation/reduction reaction at the interface, and the growth mode at room temperature is clearlyVW-like on TiO 2(1 1 0) and all the other faces studied [18]. Nucleation on the ®vefold coordinated Ti atoms was inferred from experimental and theoretical results [72,436]. The growth and nucleation behavior on (1 Â 1) and

(cross-linked) (1 Â 2) TiO2(1 1 0) surfaces was compared in a recent STM study [437]. On the (1 Â 1) surface, clusters nucleate randomlyon terraces with no apparent preferred nucleation site. On the (1 Â 2) surface, however, theygrow preferentiallyon top of the bright rows in STM. These clusters are also thermallymore stable than on the (1 Â 1) surface. On vicinal surfaces with co-existent (1 Â 2) and (1 Â 1) termination, Pt clusters attach mostlyto the end of the (1 Â 2) rows [437]. The interaction of Pt with the reduced Ti at the (1 Â 2) surface (assumed was the added row structure in Fig. 18c) is thought to account for changes in the nucleation behavior [437].

The electronic structure of Pt-modi®ed TiO2(1 1 0)-(1 Â 1) surfaces was mapped out with photon energy-dependent and angular-resolved photoelectron spectroscopy [438]. UPS measurements indicate no substantial hybridization between Pt and TiO2 states, and verylittle charge transfer on stoichiometric surfaces [72]. Electronic charge transfer was found between the Ti3 ions of pre-reduced surfaces and 140 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 47. Experimental results on an SMSI model system, Pt/TiO2(1 1 0). (A) Low-energyHe ion scattering (LEIS) spectra of (bottom) the clean TiO2 (1 1 0) surface, (center) after evaporation of 25 ML Pt at room temperature, and (top) after high- temperature treatment caused encapsulation. (B)±(F) STM and STS results after the high-temperature treatment. (B) Overview (2000 Ð Â 2000 AÊ ). Clusters are approximately200 AÊ wide and 40 AÊ high. Most clusters show hexagonal shape elongated along the substrate [0 0 1] direction (type A). A few square clusters (type B) are seen. (C) Small-scale image (500 Ð Â 500 AÊ ), ®ltered to show the structure of the encapsulation layer and the substrate. (D) Atomic-resolution image of an encapsulated hexagonal `type A' cluster. (E) Atomic-resolution image of a square `type B' cluster, showing an amorphous overlayer. (F) STS of the different surfaces. From Dulub et al. [20]. # 2000 The American Physical Society.

Pt [72]. The electrical properties of Pt/TiO2(1 1 0) were also investigated [439], and the formation of a Schottkybarrier was deduced from STS I±V curves. Electron scattering techniques showed that the Pt clusters grow with their (1 1 1) face parallel to the substrate [18,440]. Mild annealing causes a sintering of the Pt particles. However, while STM images of annealed Pt clusters show that the majorityof clusters are quasi-hexagonal (with an elongation along the [0 0 1] direction probablycaused bydiffusion effects), a minorityof square clusters was also observed (Fig. 47). These probablyhave their (1 0 0) face parallel to the TiO 2(1 1 0) substrate. Low-energyion scattering experiments clearlyshow that UHV annealing at higher temperature causes encapsulation (Fig. 47A), and glancing-exit XPS identi®ed the layer as highly reduced with (probably) Ti2 species present [18]. No evidence for encapsulation was found in a studythat employed verysimilar techniques [72]. In a convincing experiment byChambers' group it was shown that the reduction state of the substrate in¯uences whether or not heat-treatment leads to an encapsulation of the

Pt overlayer. On stoichiometric Nb-doped TiO2(1 0 0) ®lms, no encapsulation was observed. Platinum overlayers on conventionally reduced TiO2 substrates did encapsulate at similar temperatures [190]. U. Diebold / Surface Science Reports 48 (2003) 53±229 141

The overlayer in the SMSI state was imaged successfully with STM by Dulub et al. [20,71], see Fig. 47B±E. Most clusters (type A) have a hexagonal shape elongated along the substrate [0 0 1] direction and are, on average, 40 AÊ high and 200 AÊ wide. A few have a square shape (type B). Those are smaller. A simple calculation (taking into account the deposited amount of Pt (25 ML), the surface coverage after encapsulation (40%), and the cluster height) shows that the clusters resemble `icebergs' reaching several tens of AÊ ngstroms deep into the substrate. The 500 Ð Â 500 AÊ large image in Fig. 1c is ®ltered to visualize the 3D structure of the encapsulated clusters and the substrate. The tops of the clusters are smooth and ¯at, which is essential for atomicallyresolved STM measurements. (The Pt particles remind of the `pillbox' shape observed in HREM images of a catalyst in the SMSI state [441].) On top of the type A clusters, striped `zigzags' are visible. On different clusters the stripes are oriented either parallel to the substrate [0 0 1] direction or rotated by Æ608. No clear preference for any rotational orientation was observed, nor a strict correlation between the directions of stripes and cluster elongations. Fig. 47D is an atomicallyresolved image of a typeA cluster surface. The stripes are approximately15 AÊ wide and consist of bright spots arranged in a hexagonal symmetry with a distance of 3 AÊ . The bright zigzag rows contain either 5 or 6 atoms along the close-packed directions and separate triangular areas consisting of 10 atoms. Surfaces of type B square clusters exhibit no apparent long-range order (Fig. 47E) with strings of 3±6 atoms oriented along the substrate [0 0 1] direction. The `type B' clusters are probably crystallites with their (1 0 0) face parallel to the substrate. It is conceivable that such surfaces should be found at the sides of the hexagonal, (1 1 1)-oriented `type A' clusters. STM current vs. voltage (I±V) curves have been taken from the clean sputter-annealed

TiO2(1 1 0) surface, from TiO2(1 1 0) between encapsulated clusters, and from clusters of clean and encapsulated Pt (Fig. 47F). There is almost no difference in the I±V curves from clean TiO2 and from TiO2 between encapsulated clusters. The electronic structure of the substrate is not stronglyaffected by the encapsulated clusters. The electronic structure of the clusters changes from those typically observed for small metal clusters on TiO2 [162] (``clean Pt'') to a more semiconductor-like behavior after encapsulation. An atomic model of the zigzag structure in Fig. 47C was proposed in [20] and re®ned with DFT calculations in [302]. It consists of a TiOx bilayer ®lm on Pt(1 1 1). The preferred interface with Pt has Ti with O as an overlayer that stays under considerable stress. It consists of a series of linear mis®t dislocations at the relativelyweak Ti/Pt interface that are 6/7 Ti/Pt rows wide. In addition, strong interactions at the O/Ti interface and O-layer strain also cause the Ti/Pt interface to abruptly change from hcp- to fcc-site Ti, producing linear dark stripes in Fig. 47C. Alternating hcp- and fcc-site triangles, each with 10 O-atoms, are separated bybridging O in an abrupt O/Ti mis®t dislocation, thus producing a zigzag pattern. However, in the computations this structure is onlystable if the corners of the zigzag consist of Ti atoms instead of O atoms. While ®lled-state STM images indicate the ends might be different than the line portions of the zigzag features, ®rm experimental evidence for such single Ti atoms is not yet available. Adsorption of CO modi®ed the growth of metallic Pt clusters, with more `wetting' when evaporation was performed in a background of CO [442]. The substrate structure, i.e. (1 Â 1) vs. (1 Â 2) termination, was found to be in¯uential for the surface chemistryof supported Pt clusters [437]. The adsorption of H2O on model clusters was investigated using the Gaussian 90 (ab initio) program [443], and Pt was found to promote the H2O adsorption process and hole trapping. However, experimentally an enhancement in the CO photo-oxidation reaction rate or yield is not observed in the presence of Pt metal clusters on the TiO2(1 1 0) surface [444] under UHV conditions. 142 U. Diebold / Surface Science Reports 48 (2003) 53±229

TiO2(1 0 0) and (0 0 1). The modi®cation of TiO2(1 0 0)-(1 Â 3) surfaces was studied in a series of papers bySchierbaum et al. [445]. Photoemission spectroscopyof the as-evaporated ®lm indicated the presence of small Pt clusters. The position and intensityof the Ti 3 feature of the (1 Â 3) surface (see Section 2.3.2) was not altered signi®cantly, and electronic charge transfer was virtually absent. Angle- resolved XPS measurements indicated a (1 1 1)-oriented overlayer. For (0 0 1)-oriented substrates the modulations in the XPD pattern were too weak for a reliable structural analysis. Anatase (0 0 1)-(1 Â 4). On the highlycorrugated anatase (0 0 1)-(1 Â 4) surface, Pt forms 3D clusters with a narrow size distribution on the terraces. Interestingly, Pt clusters migrated along the rows without coalescing to step edges via a cluster diffusion mechanism upon annealing in vacuum. Time- dependent autocorrelation analysis revealed that the diffusion occurred preferentially along the atomic rows [446].

4.2.26. Copper The growth mode of Cu is clearlyVW-like, as seen in LEIS measurements [447] and, more recently, STM [159,295]. 3D islands nucleate preferentiallyat step edges, with some nucleation occurring also at the terraces of the (1 Â 1) surface [295], see Fig. 48. The islands have a narrow size distribution in the coverage regime displayed in Fig. 48, and the densityrather than the island size increases with coverage. For higher coverages, the islands grow mainlyin height. The apparent limitation on the island sizes for coverages below 0.5 ML is referred to as `self-limiting' island growth byChen et al. [295], and two possible explanations were proposed to account for this effect. In one case the rate at which

Fig. 48. STM images 1000 Ð Â 1000 Ð for various Cu coverages on TiO2(1 1 0)-(1 Â 1), dosed at room temperature: (a) 0.03 ML, (b) 0.13 ML, (c) 0.25 ML, and (d) 0.5 ML. The islands nucleate preferentiallyat step edges. From Chen et al. [295]. # 2000 Elsevier. U. Diebold / Surface Science Reports 48 (2003) 53±229 143 adatoms attach to existing islands drops rapidlyas the island size increases. It was speculated that this could be due to strain ®elds that accommodate the lattice match between Cu and TiO2(1 1 0). Adatoms rejected bythe islands are then available for new islands. Alternatively,the rate at which adatoms reach existing islands drops rapidlyas the Cu coverage increases; possiblydue to continued nucleation of islands at defects at the TiO2 surface. Because of the high mobilityof Cu on TiO 2, aggregation was even observed at low temperatures [447]. A lower bound for the diffusion constant of 4 Â 10À20 cm2/s was estimated [295]. After evaporation on a hot surface, or after annealing of ®lms deposited at room temperature, only3D islands were observed with STM. These islands were much larger, were still mainlylocated at step edges, and exhibited a more regular, faceted shape. Again a self-limiting behavior was observed, but with the island size depending on temperature. The temperature-dependent growth morphologyof Cu ®lms on

TiO2(1 1 0) was also studied byCarroll et al. [448]. Thicker Cu ®lms exhibit a (1 1 1) orientation on TiO2(1 1 0) [447,449,450]. Detailed HREM images of the interface were given in [450]. Atomicallysharp interfaces with no mis®t dislocations were observed. The h110i direction of the Cu overlayer is aligned with the substrate's [0 0 1] direction. The overlayer is commensurate with 1 10 but incommensurate with [0 0 1]. Two equivalent domains, rotated by180 8, are possible and were observed experimentally [379,450]. Theygive rise to {1 1 1} stacking faults and microtwins which mayoccur as a result of coalescence of 3D islands with increasing ®lm thickness. The structure of small Cu clusters was tested with FT-RAIRS with CO as a probe molecule. Two principal IR absorption bands at 2071 and 2094 cmÀ1, corresponding to CO adsorption on Cu(1 1 1)-like and Cu(1 1 0)-like sites, were found [451].

The geometryof the interface was studied with SXRD [107]. Compared to an uncovered TiO2(1 1 0) surface (Fig. 7, Table 3), the Ti atoms de-relax to their bulk-terminated positions. Large vertical and lateral displacements of oxygen atoms were observed, suggesting signi®cant bonding between Cu and O. Despite this interaction between O and Cu, no sign of an overlayer oxidation/substrate reduction was observed with photoemission and inverse photoelectron spectroscopy [396,447]. The height of Cu clusters in STM varied with the applied bias voltage [448]. This effect was used to quantifyheight variations in the Schottkybarrier formed between Cu and TiO 2. A Cu overlayer surface state was found in angle-resolved photoemission spectroscopyfor a 17 AÊ thick Cu ®lm [452]. Compared to a Cu(1 1 1) surface this Shockleystate was broadened and shifted and it exhibited a noncircular E vs. kk dispersion around G. The energyshift was interpreted as an effect of the surface step densitywhile the kk asymmetry was attributed to the state to the nature of the overlayer growth which is only commensurate with the substrate along the GM Brillouin zone direction.

4.2.27. Silver

The growth of Ag clusters was unequivocallydetermined as 3D island growth on TiO 2(1 1 0)-(1 Â 2) and (1 Â 1) surfaces [308,453±456]. (Based on LEIS measurements, a 2D island growth was inferred for deposition of Ag at 125 K [457].) The islands nucleate along step edges on both surfaces rather than at ¯at terrace defect sites. For the same coverages, the silver clusters are bigger than Cu [456]. This is in agreement with the smaller af®nityof Ag towards oxygenas compared to Cu (see Table 5) which implies a weaker interaction with the substrate. The cluster densityis larger on the (1 Â 2) surface and the cluster diameter and height is smaller than on the (1 Â 1) surface. No chemical reaction was observed for neither a stoichiometric nor slightlysputtered surfaces, but XPS showed a strong peak 144 U. Diebold / Surface Science Reports 48 (2003) 53±229 shift for verysmall clusters. According to the STM measurements of Chen et al. [456], Ag clusters deposited at room temperature are remarkablystable upon annealing to 900 K. When the Ag overlayer was deposited at 100 K, strong sintering effects occurred upon annealing to 300 K [453]. Ostwald ripening was also shown to occur upon exposure of 10 Torr O2 at room temperature [308].

4.2.28. Gold Supported Au catalysts can oxidize molecules such as CO at, or even below, room temperature.

This is all the more surprising as neither ¯at Au nor TiO2 bythemselves are particularlyactive for CO or O2 adsorption. On TiO2(1 1 0) CO adsorbs on onlyweaklyand oxygenmolecules show a strong interaction onlywith defects (see Section 5.1). The seminal work byHaruta et al. [458,459] on Au- based catalysts for combustion of CO as well as a variety of other low-temperature oxidation and hydrogenation reactions, has stimulated quite some interest in this area. The reader is referred to two review articles, one on catalysis on gold by Bond and Thompson [460], and a second one on the growth, interactions, structure, and chemistryof gold deposited on TiO 2(1 1 0), recentlypublished by Cosandeyand Madey [294]. The in depth-description in these articles goes beyond the summary provided here. It is now clear that for catalytic activity:

(1) Au must be present in the form of small particles (generallywith a diameter of <5 nm, with smaller clusters being more active, but there appears to be a minimum size for activity). (2) The Au clusters must be on the right kind of support (in the beginning it was thought that a

reducible metal oxide such as TiO2 or Fe2O3 is necessary, but in the meanwhile other supports such as SiO2, MgO, or ZrO2 have shown to be appropriate as well). (3) The preparation technique plays a big role, with `deposition±precipitation' (a precursor to the Au is brought out of solution in the presence of a suspension of the support which acts as a nucleation agent) being primarily employed by the Haruta group and leading to a catalyst with small, active Au clusters.

The interrelated issues (1)±(3) raise a number of questions. For example, the fact that the clusters need to be small has several consequences. Theycertainlycontain a number of Au atoms in undercoordinated sites, i.e., at edges and corners, and these special Au atoms might be relevant for activation of oxygen. (There seems to be an agreement that O2 activation is the rate-limiting step for CO activation, with CO adsorbing/desorbing reversiblyfrom both the Au cluster and the support.) A `quantum size effect' might be important, i.e., the clusters are so small that do not exhibit metallic character which can in¯uence the chemical reactivity [461]. The substrate does playa crucial role for activity, but it is unclear which one. It is surmised that the periphery between the Au cluster and the support is especiallyactive, or that the substrate itself (e.g. defects on TiO 2) activate the O2 that then migrates to the interface [462]. Also, the interaction with the substrate could change the lattice constant of Au, resulting in a strained lattice. In a related work on size-selected Au clusters on MgO [463], it has been pointed out that the Au clusters are slightlynegativelycharged and are most active when adsorbed on an F center. Some of these questions are hoped to be addressed with the model system at hand, i.e., Au deposited on TiO2(1 1 0) under UHV conditions. The growth of the metal overlayer is now well understood [464], and the (average) morphologies that evolve upon deposition at room temperature are schematically U. Diebold / Surface Science Reports 48 (2003) 53±229 145

Fig. 49. (a) Schematic representation of the change in Au cluster morphologyas a function of Au coverage on TiO 2(1 1 0), from [464]. (b) HRTEM image of a Au cluster viewed in cross-section formed after UHV deposition of 1 nm Au on

TiO2(1 1 0). The cluster has a truncated spherical shape with a small contact area and a large contact angle. The cluster size is ca. 5.8 nm. From Cosandeyand Madey [294]. # 2001 World Scienti®c. 146 U. Diebold / Surface Science Reports 48 (2003) 53±229 summarized in Fig. 49a. Verysmall coverages of Au show `true' 2D growth, i.e., one-layerhigh, small islands, with nucleation at the Ti(5) sites and defects. The kinetic limitations during growth result in a `quasi-2D' morphology, with two-layer high clusters. The critical coverage for this morphology, given in Fig. 49a, changes somewhat with temperature. These 2D clusters do not bind CO at room temperature, at least not much [464]. Theyare also not active for O 2 adsorption, but atomic oxygen is bound stronger [23] on 2D than on thicker clusters. Under these conditions, the oxidation of CO on this model system is very rapid at room temperature [23]. Such quasi-2D clusters are not yet metallic, as shown in an STS measurement byGoodman and co-workers [22]. In accompanying high-pressure studies this model system was active for CO oxidation, and the activity was related to the size of the band gap [22]. (These measurements, while suggestive to provide some of the answers to the above questions, should nevertheless be taken with some caution, as STS measurements of clusters on semiconductors are not easyto interpret.) XPS shifts of the 4f levels in the veryinitial stage (2D islands, <0.1 ML) are interpreted as ®nal-state screening effects. As expected (Section 4.1.1), the shape of the substrate core levels is not changed, but some band bending (downward) during this initial stages of growth has been attributed to charge transfer from the cluster to the substrate [465]. It should also be pointed out that sometimes a shoulder that is present in XPS Au 4f levels of `real' catalysts. However, the absence of this, possiblyoxidized, Au in the model systemmight not be of too much relevant, since it is present in activated as well as deactivated catalysts, and since UHV-deposited Au/TiO2(1 1 0) does show catalytic activity. When more Au is deposited at room temperature, the Au clusters assume a hemispherical shape. Thicker layers show a worm-like percolation network. Upon annealing, clusters irreversibly change to an equilibrium shape which is quasi-spherical for smaller clusters (Fig. 49b) and faceted for thicker ones [24,294,306,465]. The cluster depicted in Fig. 49b might give an idea of the shape of an active particle. It is of about the right size range (a mean cluster diameter of 2±3 nm is quoted for optimum activity [460], but there is always a range in size distribution). The more hemispherical clusters in an annealed Au overlayer might be more relevant as model systems, as catalysts are typically calcinated (i.e. annealed) in air at elevated temperature after impregnation. Also, STM investigations at elevated pressures and temperature were performed recentlybyKolmakov and Goodman [309,310], and change in cluster size was found for both, oxygen and CO background pressures, which also indicates that clusters will resemble more such an equilibrium shape rather than a (quasi)-2D island. However, there is an emphasis on `strong interfacial bonding' between Au and the support that is necessaryfor a high activity. Just what this strong bonding might be, and how it would affect the particle shape, is unclear.

Cosandeyet al. [464] have analyzed HRTEM images of UHV-deposited Au clusters on TiO2(1 1 0) and have found a (positive) interfacial energythat ranges around 900 mJ/m 2, depending on the size of the cluster and analysis technique used.

The epitaxial relationship between Au and rutile TiO2(1 1 0) was mapped out bythe same group [306,464]. Two kinds of orientation for (thicker) clusters have been found. In the ®rst one the

Au(1 1 1) layer is in direct contact with the TiO2(1 1 0) substrate. The close-packed Au110 direction is parallel to the substrate [0 0 1] direction, with a lattice mismatch of 2.6%. The perpendicular direction has a high lattice mismatch of 14.3%. This mismatch is reduced in the second con®guration, where Au(1 1 2) is parallel to the TiO (1 1 0) substrate, again with 110 parallel to 001 , but a 2 Au TiO2 better ®t (mismatch of 8.4%) in the perpendicular direction. The epitaxial relationship between Au and anatase is described in [466]. Clearly, the Au particles are somewhat strained, but since this mismatch is expected to be different on different substrates, it might not be a major cause for the U. Diebold / Surface Science Reports 48 (2003) 53±229 147 change of Au from a noble metal in bulk form to a catalytically active material when present as a supported cluster.

As mentioned before, the metal/support interaction is subject to much speculation. On TiO2(110) there is some indication from STM studies that the clusters nucleate at defect sites [467].These might anchor the clusters more ®rmlyto the substrate, but could also provide some excess charge to render more anionic Au [468]. A recent photoemission studyshowed that TiO 2(1 1 0)-supported Au clusters are active for SO2 adsorption [469] and DFT calculations indicated that the Au could `attract' the bulk defects in the reduced substrate towards the surface. This is an intriguing idea and it would be good if it were backed up with other experiments. As pointed out in [294,460] a whole range of unanswered questions exists. For example, how does the cluster/substrate interface look like? What are the exact atomic positions? Can a system be prepared that mimics the active catalyst? What about anatase substrates? What is the role of bulk/surface defects for preparation of the catalyst and during the reaction? Are the clusters charged? Is their surface actuallyamorphous under reaction conditions [460]? What is the catalytically active state of oxygen? Some of these questions could possiblybe addressed with a more in depth investigation of Au on single-crystalline TiO2 surfaces, ideallyhand-in-hand with high-pressure studies on model systemsas well as real catalysts.

4.3. Conclusion

The number of papers published on metal overlayer growth on TiO2 is trulyimpressive. Virtuallyall the 3d and 4d transition metals, all alkalis, and aluminum have been studied, and the most important materials of the 5d metals as well (Fig. 39, Table 6). (Interestingly, no studies of lanthanides or actinides on single-crystalline TiO2 surfaces have been performed, at least not to this author's knowledge.) It is intriguing that, with veryfew exceptions, all experimental investigations show a clear trend across the periodic table. Metals can be roughlyclassi®ed in `non-reactive' ones where no interfacial oxidation/ reduction reaction happens, and reactive overlayers, where oxygen is extracted from the substrate and is incorporated into the ®lms. The strength of this reaction scales with the heat of formation, and so does the tendencyto wet the substrate. The nucleation behavior starts to be suf®cientlywell characterized. There is a clear trend for the epitaxial relationship of groups of overlayers with the substrate, and a predictive understanding of the thermal stabilityof the ®lms has been obtained.

While the trends of metal/TiO2 growth are now rather well understood, comparablylittle effort has concentrated on the growth of metal oxides on TiO2, see the relative scarcityof such growth studies in Table 6. While reactive overlayers become oxidized on TiO2 theydo so on the cost of TiO 2 reduction. The interfacial reactions and the resulting overlayer properties might be quite different when an additional supplyof oxygenis offered from the gas phase. Although details are still missing for some elements, the rich bodyof experimental data on metals/

TiO2 as a whole constitutes a systematic and conclusive data base. It should be a rewarding challenge for theorists to follow up with state-of-the-art calculations in order to provide insight that cannot be derived from experimental work alone. It is comforting to see that such works have started to appear. Adsorption on these rather well-characterized systems has also started to stimulate interest. Such studies should provide a next step in order to trulyunderstand the promotional effects of supported metal catalysts. Finally, it will be interesting to see which of the concepts derived from the already rather complete work on metals on TiO2 will be transferable to other oxide supports. 148 U. Diebold / Surface Science Reports 48 (2003) 53±229

5. Surface chemistryof TiO 2

The adsorption of molecules and atoms, and their dissociation and/or reaction to other products, is certainlythe most extensive area of studyin the surface science of TiO 2. General considerations about different types of adsorption mechanisms at oxide surfaces are found in [1]. The discussion in this review is split (somewhat arbitrarily) into a section on inorganic and one on organic molecules. For convenience, information about the various molecules is brie¯ysummarized in table format. It should be pointed out that recent structural investigations have rendered some surprises about the surface geometrical structure and morphologyof TiO 2 surfaces (see Section 2) which might have affected results on molecular adsorption and surface chemistry. This is pointed out in cases where it is obvious that details of the surface structure of TiO2, which were previouslyunknown, mayhave in¯uenced the results. For the most part, however, the results have not been re-interpreted and the view of the surface structure was adopted that was given in the original articles.

5.1. Inorganic molecules

5.1.1. Hydrogen

The interaction of TiO2 with hydrogen at higher pressures is interesting from a technological point of view. Pd-sensitized TiO2 is used as a hydrogen sensor [470] and reducing oxide powders in a H2 atmosphere is typically employed in catalysis [471]. UHV studies have consistentlyshown that TiO 2 5 surfaces do not interact stronglywith molecular hydrogen [472]. High doses of H2 (10 L) at room temperature caused additional emission peaks in the valence band region in UPS [27].In[131] it was argued that oxygen vacancies act as special adsorption sites for hydrogen. In contrast to molecular hydrogen, atomic H sticks to TiO2(1 1 0) surfaces at room temperature [127]. Despite the weak interaction between molecular hydrogen and TiO2, there are indications of a reduction when a TiO2 surface is annealed in a H2 atmosphere under high vacuum conditions [473,474]. It is possible that so-called `clean' single-crystalline surfaces, obtained with the usual cleaning procedures, are to some extent covered byhydrogen(see also Section 2.2.1.4). The extent to which this occurs, as well as the form and the origin of the hydrogen, is still being debated. This adsorbed hydrogen could come from two sources. Firstly, it could come from the water in the residual gas pressure in the UHV chamber. The high sticking coef®cient of H2O together with the tendencyto dissociate at vacancies (see Section 5.1.2) can quickly lead to a hydroxylation of all oxygen vacancies.

Secondly, hydrogen from the bulk of TiO2 crystals could end up at the surface. With RBS it was determined that a stoichiometric TiO2 sample contains ca. 1 at.% hydrogen [475]. In a recent ion scattering/ion-induced recoil study, a TiO2(1 1 0) was stepwise annealed to higher temperature. The O and H content was monitored during the heating excursion. The studycomes to the surprising conclusion that all bridging oxygen atoms of a UHV-annealed surface are hydroxylated, even after annealing to 730 8C [476]. It needs to be pointed out, however, that this TiO2 sample was not treated prior to the heating excursion. It is doubtful that such a high coverage with hydrogen is also found when a crystal is prepared by the usual sputter/annealing cycles in UHV (Table 7).

5.1.2. Water

For manyreasons water is probablythe most important adsorbate at TiO 2 surfaces. Manyof the applications mentioned in Section 1.2, for example, almost all photocatalytic processes, are performed U. Diebold / Surface Science Reports 48 (2003) 53±229 149

Table 7

Surveyof hydrogenadsorption on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference Rutile (1 0 0)-(1 Â 3) UPS, ELS, AES [27]

Rutile (1 1 0), (1 0 0), (0 0 1) UPS, TiO2 surfaces do not interact stronglywith hydrogen [759] Rutile (1 1 0) LEED, XPS, ELS, EPR, work function, surface conductivity [511]

Rutile (1 1 0) LEIS, molecular H2 does not adsorb at room temperature, [127] atomic H sticks at room temperature

Rutile (1 1 0) STM, O±H and Ti±H species on slightlydefective TiO 2(1 1 0) [133] Rutile (1 1 0) CAICISS, TOF-ERDA, 1 ML of H even after annealing to 730 8C [476]

in an aqueous environment. Water vapor in the ambient interacts with TiO2 surfaces, and surface hydroxyls can easily affect adsorption and reaction processes. Water is one of the main components in the residual gas in UHV chambers, hence it is an important adsorbate even in well-controlled experiments. The adsorption of water on TiO2 has been of intense interest in recent years. It has been investigated with a varietyof experimental and theoretical techniques, see Table 8. Recently, an excellent review article on water adsorption on solid surfaces was given byHenderson [477]. This work represents a substantial expansion from a previous review byThiel and Madey [478] on the same subject. Because Henderson's article refers extensivelyto single-crystallineTiO 2 surfaces, the discussion on water TiO2 is kept brief here. For an overview, most of the recent literature is summarized in Table 8, and a brief summary, especially on unclari®ed questions regarding adsorption on the

TiO2(1 1 0) surface, is given in the following. For manyof the details, the interested reader is referred to Henderson's article [477]. Rutile (1 0 0). Experimental and theoretical studies for the most part agree that water can dissociate to some extent on perfect rutile (1 0 0) surfaces. Except for some earlywork [27], results from different spectroscopies generallyindicate that water adsorbs both dissociativelyand molecularly [174,479± 484]. This is independent of steps, point defects [479±481], and the Ti3 sites present at the (1 Â 3)- reconstructed surface. The amount of dissociated water is slightlydifferent at the (1 Â 1) and (1 Â 3)- reconstructed surface [197]. The sticking coef®cient is unityat a sample temperature of 130 K [197]. Adsorption/desorption occurs reversibly, i.e., molecular water is the only desorption product, and the surface generallydoes not become oxidized. It has been reported that veryhigh doses of water vapor or exposures to liquid water [484] causes oxidation of defects, but these results have not been reproduced in other studies [92,485]. Most theoretical results [214,241,486], but not all of them [105], agree with the notion of initial dissociative adsorption, followed bymolecular adsorption at higher coverages. Rutile (1 1 0). Despite the extensive work on water adsorption on rutile (1 1 0), see Table 8, there is a considerable disagreement on the initial adsorption behavior of water, especiallybetween theoretical and experimental studies. While most of the experimental results agree that H2O does not dissociate on TiO2(1 1 0), except at defect sites, most theoretical studies predict dissociative adsorption. Brinkleyet al. [487] performed a molecular beam scattering studywhere theyinvestigated the dynamical adsorption properties of water. A sticking probability of unity at all coverages and temperatures up to 600 K was found. The water molecule is trapped, with enough time to sample the potential energysurface of TiO 2(1 1 0) before desorption. It was also found that veryfew of the molecules dissociate, even in the limit of zero coverage. This is in agreement with TPD studies, most notablythe ones published byHenderson [128,174] and Hugenschmidt et al. [175]. A high-temperature 150 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 8

Surveyof water adsorption on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference Review of water adsorption on solid surfaces (includes [477]

extensive surveyof water/TiO 2) Rutile (1 0 0), experiment Rutile (1 0 0)-(1 Â 3), sputtered UPS, ELS, AES, TPD, water is adsorbed molecularlyon [27] 3 and with Ti overlayer TiO2(1 0 0)-(1 Â 3), but dissociates on Ti achieved through different surface pretreatments Rutile (1 0 0)-(1 Â 1), -(1 Â 3); UPS, water adsorption independent of steps and O vacancies, [479±481] with point defects; vicinal adsorbs molecularlyat 130 K and dissociates to form OH at 293 K surface Rutile (1 0 0)-(1 Â 1), -(1 Â 3), TPD, unitysticking probabilityat 130 K, dissociative and molecular [197]

sputtered states, desorbs exclusivelyas H 2O, water more weaklybound to the sputtered surface Rutile (1 0 0) TPD, isotopic labeling, dissociative desorption; recombinative [174] desorption (see Fig. 51) Rutile (1 0 0) SIMS, traces of F enhance hydroxylation of the substrate [482,483]

Rutile (1 0 0) XPS, UPS, H2O possiblyheals vacancies created bye-beam [484,760] and Ar bombardment Rutile (1 0 0), theory Rutile (1 0 0) Cluster calculations, dissociation of water [486] Rutile (1 0 0) Hartree±Fock dissociative adsorption [214] Rutile (1 0 0) Electronic structure calculations, dissociative adsorption is [241] thermodynamically favored Rutile (1 0 0) SCF MO LCAO calculations, dissociative adsorption [761] Rutile (1 0 0) Carr±Parinello MD simulation molecular adsorption on stoichiometric [105] surface, spontaneous dissociation of water at an O vacancy Rutile (1 1 0), experiment Rutile (1 1 0), Ar-bombarded UPS, dissociative adsorption at low coverages [762] Rutile (1 1 0), Ar and e-beam UPS, HREELS, H2O does not adsorb at stoichiometric surfaces at [763] bombarded room temperature, dissociative adsorption at defects Rutile (1 1 0) with vacancies Photoemission, dissociative adsorption on vacancies and on regular [489] lattice sites Rutile (1 1 0) LEIS, isotopic labeling, water adsorbs at vacancysites [127] Rutile (1 1 0), defects XPS, high exposures needed to heal electronic defect state [764] Rutile (1 1 0) TPD, XPS, workfunction, ®rst layer adsorbs at ®vefold Ti4 sites; [175] distinct second layer state at 180 KÐ`bilayer' structure with water molecules lying ¯at Rutile (1 1 0) HREELS, TPD, isotopic labeling, water adsorbs dissociativelyfor [128,174] exposures up to 7 Â 1013 mol/cm2; molecular for higher coverages; ideal surface not active for water dissociation (see Fig. 51), second water layer interacts mainly with bridging O sites Rutile (1 1 0), (1 Â 1), (1 Â 2), Molecular beam scattering, TPD sticking coef®cient of water 1up [487] and sputtered surface to 600 K; long residence time before desorption; onlyfew of water molecules dissociate

Rutile (1 1 0) TPD, dissociation of D2O and production of D2 on defects created [490] bythermal annealing, healing of vacancies Rutile (1 1 0) STM, terminal OH groups at Ti4 sites at step edges [491] Rutile (1 1 0) STM, photoemission, water adsorbs molecularlyat 150 K, dissociates [488] at 290 K, strong lateral interaction leads to formation of water `chains' U. Diebold / Surface Science Reports 48 (2003) 53±229 151

Table 8 (Continued ) Substrate Techniques/adsorption/reaction Reference Rutile (1 1 0) STM, dissociation at point defects [134] Rutile (1 1 0) MIES, UPS, dissociative adsorption at defects followed by [765] multilayer adsorption; strong lateral interactions between adsorbed water molecules Rutile (1 1 0), theory

Rutile (1 1 0) Cluster calculations, TiO2(1 1 0) is catalytically active for water [486] dissociation Rutile (1 1 0) Hartree±Fock dissociative adsorption [214] Rutile (1 1 0) Self-consisted tight-binding embedded clusters, electronic [766,767] structure of hydroxylated surface Rutile (1 1 0) DFT calculations, dissociative adsorption is most stable [768] Rutile (1 1 0) Cluster calculation, dissociative adsorption is favored [215] Rutile (1 1 0) SCF MO LCAO calculation, dissociative adsorption [769] Rutile (1 1 0) DFT embedded cluster calculations [242,518]

Rutile (1 1 0) DFT calculations of dissociated H2Oin(1Â 2) geometry [770] Rutile (1 1 0) DFT calculations, dissociative adsorption [102] Rutile (1 1 0) First-principles MD calculations, dissociative adsorption [492] Rutile (1 1 0) DFT and HF calculations, dissociative at low coverage, [99] molecular and dissociative at higher coverages Rutile (1 1 0) DFT calculations, dissociative adsorption at lower coverage, [493] molecular adsorption stabilized through intermolecular H bonding Rutile (1 1 0) SCF MO LCAO calculation, molecular adsorption stabilized [761] byH bond formation Rutile (1 1 0) Electronic structure calculations dissociative adsorption is [241] thermodynamically favored, potential barrier towards proton transfer Rutile (1 1 0) Embedded cluster calculations, molecular adsorption is energetically [494] favored

Rutile (1 1 0) Carr±Parinello MD simulation perfect TiO2(1 1 0) surfaces adsorb [105] undissociated water, no spontaneous dissociation at O vacancies Rutile (1 1 0) DFT calculations, on perfect surface: dissociation is endothermic [134] at low coverages, weaklyexothermic at high coverages; at defects: dissociation is exothermic Rutile (0 0 1), experiment Rutile (0 0 1) UPS, dissociative adsorption at 250 K, independent of Ti3 states; [771,772]

desorption as H2O Rutile (0 0 1) UPS, AES, LEED, dissociative adsorption only [773] Rutile (0 0 1) XPS, on stoichiometric and reduced surfaces [473,774] Anatase Anatase (1 0 1), (0 0 1) DFT calculations, (1 0 1): molecular adsorption on Ti(5); hydrogen [70] atoms of the water molecule align towards bridging oxygen atoms (0 0 1): dissociative adsorption up to 0.5 ML Anatase (1 0 1) First-principles MD calculations molecular adsorption [553] Anatase (1 0 1) TPD and XPS, exclusivelymolecular adsorption [495] Anatase (0 0 1) Hartree±Fock [214] Anatase (0 0 1) Cluster calculations, dissociative adsorption is favored [215] Anatase (0 0 1) Water co-adsorption with oxygenÐsee Table 9 [215]

Anatase (0 0 1) Water co-adsorption with CO2Ðsee Table 10 [215] Anatase (0 0 1) Water co-adsorption with NH3Ðsee Table 12 [215] 152 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 8 (Continued ) Substrate Techniques/adsorption/reaction Reference Liquid water/wettability Rutile (1 1 0), (0 0 1), (1 0 0) Contact angle measurements, XPS reversible wettability [92,485] with UV-created defects (see Fig. 52)

Rutile (1 1 0) SHG of H2O/TiO2(1 1 0) interface, UV-generated surface defects [775] Rutile (1 1 0) X-raystanding waves, Rb and Sr ion distribution in electric double layer [776] Anatase (1 0 1), (0 0 1) LEED, bulk-terminated surfaces, impedance spectroscopymeasurements [777] of flatband potential tail in TPD spectra was attributed to recombinative desorption of water that has dissociated at defects, i.e., oxygen vacancies [128,174,175]. These authors conclude that the ¯at, unperturbed TiO2(1 1 0) surface does not lead to dissociation of the water molecule, but that water dissociates at point defects. Most of the photoemission and other spectroscopic experiments listed in Table 8 are consistent with this idea, although there is some deviation in the details. For example, there is disagreement about the temperature at which water dissociates at point defects [128,488,489]. Also, while most authors report reversible adsorption, i.e., onlyH 2O leaves the surface upon heating, some studies [490] report the production of hydrogen molecules at point defects, combined with a healing of these defects [487].It was pointed out [170,477] that the different preparation techniques of TiO2 surfaces that can lead to drasticallydifferent surface morphologies and a range of undercoordinated sites (see Figs. 18 and 21) could possiblybe a reason for the observed differences. Recent STM studies [134,488,491] support the earlier experimental results that dissociative adsorption occurs at defects, and that water molecules stay intact when adsorbed at ¯at terraces.

The difference between the adsorption behavior of water at the TiO2(1 1 0) and (1 0 0) surfaces, exempli®ed in Fig. 50 is rationalized with the structural model proposed byHenderson ( Fig. 51). Water

Fig. 50. Comparison of TPD spectra obtained from water adsorption on TiO2(1 1 0) (dashed line) and TiO2(1 0 0) (solid line). The water adsorption temperatures were 140 and 138 K, respectively. The traces exhibit signi®cant differences in the number of monolayer desorption states and the relative amount of water desorbing above 300 K. Isotopic labeling studies indicate molecular adsorption on TiO2(1 1 0), except for adsorption at defect sites which gives rise to a high-temperature peak (not shown in these traces). The TiO2(1 0 0) surface dissociates water. From Henderson [174]. # 1996 The American Chemical Society. U. Diebold / Surface Science Reports 48 (2003) 53±229 153

Fig. 51. Schematic model for the interaction of water with the (1 1 0) and (1 0 0) surfaces of TiO2 in side and perspective 4 view. Water binds at the acidic sites (the ®vefold coordinated Ti ions) on TiO2 with the O±H bonds pointing awayfrom the Ê surface. On TiO2(1 1 0), the distance between the water molecule and the next bridging oxygen atoms is 3.2 A, precluding a hydrogen-bonding interaction between the adsorbed water molecule and the bridging oxygen atoms. On TiO2(1 0 0), the O±O distance between an adsorbed water molecule and a twofold coordinated O atoms is smaller, favoring hydrogen bonding and dissociation. From [174]. # 1996 The American Chemical Society. is expected to adsorb at exposed ®vefold coordinated Ti sites in each case with the hydrogen atoms pointing awayfrom the surface. Since (multiple) hydrogenbonding interactions between the adsorbed species and the bridging oxygen atoms of the substrate are expected to facilitate proton transfer, the relative distance between the water molecule and the neighboring bridging oxygen atoms is of importance. On the TiO2(1 0 0)-(1 Â 1) surface, the molecule adsorbs in an inclined fashion, and the distance between the oxygen atom in the water molecule and the twofold coordinated substrate O2À ions is small enough for a weak H±O interaction to occur. In contrast, the distance between an O atom 2À Ê adsorbed at a Ti atom and a neighboring bridging O ion is more than 3 A on TiO2(1 1 0), precluding O±H interaction. This model would also explain whyoxygenvacancies on TiO 2(1 1 0) are particularly reactive towards H2O dissociation. An H2O molecule adsorbed in a vacancywould provide a geometricallyparticularlywell-suited adsorption site for O±H interactions and dissociation. Wang et al. [92,485] pointed out that water dissociation at these oxygen vacancies possibly has a macroscopic effect on the wetting abilityof water, see Fig. 52. Generally, TiO2 surfaces are oleophilic and hydrophobic, i.e., water does not wet the surface but oil does. However, after a TiO2 sample was exposed to UV light, the contact angle of water droplets decreases to essentiallyzero degrees [92,485]. Storing such amphiphilic surfaces in the dark restores the hydrophobicity of the original surface. The surface stays oleophilic at all times, i.e., it exhibits `amphiphilicity'. Based on UHV studies, this is attributed to the creation of surface vacancies that dissociate water and form microscopic hydrophilic domains [92,485]. This effect is exploited in the production of antifogging and self-cleaning coatings on mirrors.

In stark contrast to the experimental results on H2O/TiO2(1 1 0), most theoretical calculations indicate that dissociation of the water molecule is energeticallyfavored on the perfect TiO 2(1 1 0) 154 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 52. Photographs from a TiO2(1 1 0) single crystal. (a) The surface is generally hydrophobic, and water droplets do not wet. (b) After ultraviolet irradiation, the surface becomes hydrophilic, presumably because of the formation of oxygen vacancies which facilitate the dissociation of water. (c) Exposure of a hydrophobic TiO2-coated glass to water vapor. The formation of fog (small water droplets) hinders the view of the text on the paper placed behind the glass. (d) Creation of an antifogging surface byultraviolet irradiation. The high hydrophilicityprevents the formation of water droplets. The TiO2 surface stays oleophilic, irrespective of the treatment. From Wang et al. [92]. # 1997 MacMillan Magazines Limited. surface (see Table 8). Intermolecular H bonding was invoked to act as a stabilizing factor for a mixed dissociated/molecular state [99,492,493]. Other calculations point out that, while dissociative adsorption on rutile (1 1 0) is thermodynamically favored, it might be hindered by a potential barrier [241]. One cluster calculation showed molecular adsorption [494], while manyothers, using a similar approach, predict dissociative adsorption, see Table 8. A veryrecent DFT slab calculation byNorskov and co-workers [134] agrees with the experimental data (water dissociation is an endothermic and exothermic process on terraces and point defects, respectively), and it is argued that the con®guration of the water molecules in test geometries plays an important role for the calculated energetics. A molecular dynamics simulation using the Carr±Parinello approach [105] found molecular adsorption on stoichiometric TiO2 surfaces. However, these calculations do not reproduce other experimental results. Placing an H2O molecule in a bridging oxygen vacancy did not lead to spontaneous dissociation on the TiO2(1 1 0) surface. Also, in [105] it was concluded that single OH groups are not stable on TiO2(1 1 0), in contradiction to recent STM data [134,488]. U. Diebold / Surface Science Reports 48 (2003) 53±229 155

The adsorption behavior of thicker H2O layers was investigated by several groups [128,174, 175,197,477,487].OnTiO2(1 1 0), a distinct second layer peak at 180 K is observed in TPD. This was interpreted as a `bilayer' structure, where the second water layer lays ¯at [175] and, based on additional HREELS measurements, it was concluded that water in this second layer interacts onlywith the bridging oxygenatoms [128]. Water multilayers on TiO2(1 0 0) (which partially dissociates water, see above), followed a zeroth-order desorption behavior. This is in contrast to multilayers on TiO2(1 1 0), where deviations from this desorption behavior were found [128,174, 197,477]. The adsorption of water on anatase surfaces has so far been mostlystudied theoretically(see Table 8). On the basis of DFT and ®rst-principles molecular dynamics calculations, Selloni and co-workers [70] concluded that water adsorbs molecularlyon anatase (1 0 1). A TPD/XPS studybyHerman et al. [495] con®rms this prediction. Calculations of water adsorption on the anatase (0 0 1) surface [70,214] predict dissociative adsorption but have yet to be con®rmed by measurements.

5.1.3. Oxygen Because Ti is such a reactive element, oxygen-de®cient surfaces are clearly expected to react with

O2. In manystudies it has implicitlybeen assumed that oxygenexposure would just ®ll surface vacancies on TiO2. It is onlyrecentlythat the intricacies of the oxygen/defect interaction was investigated in more detail. (In this context, the reader is reminded that annealing reduced TiO2(1 1 0) crystals in an oxygen background pressure causes a reoxidation of the crystal that leads to a restructuring of the whole surface. This phenomenon is discussed in Section 2.2.2.2.) Photocatalytic studies [496±498] as well as co-adsorption studies of water and ammonia with oxygen-predosed surfaces [275,499,500] have revealed that the `®lling' of oxygen vacancies is not as simple as previouslythought. Models for the adsorption of oxygen at low and somewhat higher temperatures were derived from these studies and are depicted in Fig. 53. Oxygen does not adsorb at

100 K to a stoichiometric TiO2(1 1 0) surface. When a surface with 8% vacancies is exposed to O2 at cryogenic temperatures, the saturation coverage is about three times the vacancy concentration [275]. À This was rationalized with a model in which O2 species are bound in the vicinityof a vacancy,see Fig. 53A. Most of these oxygen molecules desorb around 410 K; irradiation with 4.1 eV photons results in photodesorption [500]. Molecular oxygen, adsorbed at 105 K, was also found to be active for the photo-oxidation of CO [498]. From co-adsorption studies with water and ammonia it was concluded that O2, when dosed at a temperature between 90 and 600 K, results in the ®lling of an oxygen vacancy and an oxygen adatom [499],seeFig. 53. These adatoms lead to additional dissociation of water. Theydisappear when the surface is heated to 600 K. The change in the SHG 4 2À signal after O2 adsorption at room temperature was attributed to the formation of a Ti :O complex [501]. In this context, it should be noted that adsorption of oxygen (or another electronegative element) at ®vefold coordinated Ti sites does not violate the rules for creating a stable surface (see Section 2.2.1.1). While a compensated surface under perfect vacuum conditions will have emptyTi sites, it has been shown for several oxides that the presence of a chemical environment can result in a surface that has a higher stabilitywhen covered with an adsorbate [502,503]. In particular, this is the case for

RuO2(1 1 0), which has the same rutile structure as TiO2. Under the conditions of a high oxygen chemical potential the so-called `cus'-sites (5-fold coordinated Ru atoms) are preferablyoccupied with a singlycoordinated O atom [504] (Table 9). 156 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 53. (a) Schematic model for the bonding of O2 to the vacuum-annealed TiO2(1 1 0) surface at low temperatures. Model A shows the top view of the TiO2(1 1 0) surface with an oxygen vacancy site. Model B shows top and side (along the [0 0 1] direction) views of O2 molecules bonding at vacancyand non-vacancysites. From Henderson et al. [275]. # 1999 The American Chemical Society. (b) Schematic model for the formation of oxygen adatoms from the interaction of O2 with oxygen vacancies at higher temperatures, and for the role of these adatoms in dissociating coadsorbed water. From Epling et al. [499]. # 1998 Elsevier.

5.1.4. Carbon monoxide and carbon dioxide

5.1.4.1. CO. Adsorption of CO on metal-promoted TiO2 surfaces was studied byseveral authors [23,395,415,430,434,505±509], and is partiallydiscussed Section 4.2 (see also Table 6). Of particular interest in this context is the low-temperature oxidation of CO on nanosized, TiO2-supported Au particles, discussed above (Section 4.2). In contrast to metal-promoted surfaces, the adsorption of CO on the clean TiO2(1 1 0) surface is far less investigated experimentally [131,510±512], although this system was treated theoreticallybya number of groups [242,513±520]. An earlystudybyGo Èpel et al. [131,511] found oxygen vacancies to be special adsorption sites for H2 and CO. de Segovia and co-workers [512] found onlya small CO coverage after exposure of a sputtered surface to 10 5 L. Nevertheless, this small amount of CO had a dramatic influence on the O ion yield in ESD. Yates and co-workers [510] performed a TPD studyof CO adsorption on stoichiometric and defective

TiO2(1 1 0) surfaces. CO was dosed at 105 K, and with increasing coverage, CO was found to desorb at 170±135 K from the stoichiometric surface, see Fig. 54. From the TPD data, an initial activation U. Diebold / Surface Science Reports 48 (2003) 53±229 157

Table 9

Surveyof oxygenadsorption on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference

O2 Rutile (1 1 0) Two adsorption states (a and b) with different photodesorption [496±498] behavior and photocatalytic reactivity, see Section 5.3.3

At 120 K three O2 molecules adsorb in the vicinityof oxygen [275] vacanciesÐsee also co-adsorption with H2O and NH3 Rutile (1 1 0) At elevated temperaturesÐre-oxidation of the reduced bulk, `restructuring', see Section 2.2.2.2 4 2À Rutile (1 1 0) SHG, XPS, O2 `heals' defects slowly, indication of a Ti :O [501] complex

Rutile (1 1 0) Carr±Parinello MD simulation; O2 dissociates in vacancyslowly [105] O2 H2O Rutile (1 1 0), with TPD, higher T:O2 dissociates at vacancies and creates O adatoms, [275,499] oxygen vacancies these facilitate dissociation; low T: molecular adsorption, O2 coverage is three times that of O vacancies (see Fig. 53)

O2 H2O Rutile (1 1 0) TPD, SSIMS, EELS, UV-illumination of O2 at vacancies leads [500] to photodesorption; photolysis of an O2±H2O adduct in thick water overlayer O2 CO Rutile (1 1 0) see Table 10 energyof 9.9 kcal/mol at zero coverage and a repulsive interaction of 2.2 kcal/mol was extracted. While this is lower than values reported in previous works [131,511,521], it is probablythe study performed under the best-characterized conditions. On the `pre-annealed' surface, containing O vacancies, CO remained on the surface up to 350 K (Fig. 54). Both surfaces, the stoichiometric and the

18 18 Fig. 54. (a) TPD spectra for CO from the fullyoxidized TiO 2(1 1 0) surface. The CO exposures displayed range from 7:1 Â 1012 to 2:1 Â 1014 molecules/cm2. (b) Enlargement of the TPD spectra to show the high-temperature CO states from the 13 18 2 oxidized and the annealed TiO2(1 1 0) surface. The CO exposure was 8:6 Â 10 CO=cm for both surfaces. The high- temperature desorption features on the pre-annealed surface are attributed to CO interaction with point defects. From Linsebigler et al. [510]. # 1995 The American Chemical Society. 158 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 10

Surveyof the adsorption of C-containing inorganic molecules on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference CO Rutile (1 1 0) XPS, AES, LEED, ELS [131,511] Rutile (1 1 0) ESD, AES, a high dosage of CO at sputtered [512,549] surface produces onlya small coverage Rutile (1 1 0) TPD, CO adsorbs weaklyon stoichiometric surface, [510] desorbs at 170 K (low coverage) to 135 K (higher coverage); desorbs at higher temperatures on surface with O vacancies; see Fig. 54

CO O2 Rutile (1 1 0) TPD, photodesorption, O2 blocks active site for [498,510,522] high-temperature CO desorption peak; O2 molecules at O vacancies photo-oxidize CO to CO2 CO Rutile (1 1 0) Theory: see Table 11

CO2 Rutile (1 1 0) LEED, XPS, ELS, surface conductivity, work function [131] CO2 H2O Rutile (1 1 0) TPD, SSIMS, HREELS, CO2 binds to vacancies slightly [277] 4 more strongly; linear bonding to regular Ti sites; CO2 is blocked/displaced byH 2O adsorption; formation of bi-carbonate when CO2 and H2O are dosed simultaneously CO2 H2O Rutile (1 1 0), (1 0 0) HREELS, photoreduction of CO2 in the presence of H2O [524] CO2 Na Rutile (1 1 0) Promotion with Na activates the TiO2(1 1 0) surface for [304,305,323, CO2 adsorption; formation of a CO3 complex (see Fig. 42) 325,326,328,342] slightlydefective one, show the same CO coverage at similar exposures. The saturation coverage was determined as 2:5 Â 1014 CO/cm2, corresponding to about half the number of ®vefold coordinated Ti ions. CO was the onlydesorbing product. No CO 2 production was observed. TPD measurements with isotopicallylabeled molecules showed no scrambling with substrate oxygen.When a surface with O vacancies was pre-dosed with O2 at 105 K prior to CO adsorption, the high-temperature tail in the CO TPD spectrum was suppressed, indicating that the high-temperature features in Fig. 54 are due to adsorption at defect sites [510]. Pre-adsorbed O2 molecules at the oxygen vacancies induce CO2 photoproduction [498,522] (Table 10).

The adsorption of CO/TiO2(1 1 0) was treated with different computational techniques, and a summaryof the results is given in Table 11. Adsorption with the C-end down is consistentlyfound to be energeticallymuch more favorable than the alternative orientation, where the CO molecule would bind with the O-end down. Throughout these calculations the C±O bond distances are rather similar, but the adsorption energies varyconsiderably,with the experimental value from [510] being on the lower end of the given range. The decrease in binding energywith higher coverage, caused bythe onset of repulsive interaction between CO molecules, is reproduced in slab calculations. Experimentally, the C± O stretching mode of the adsorbed CO molecule shows a red shift compared to a free molecule [523]. The calculated values of the red shifts of the C±O stretching mode varyconsiderably.A quantitative comparison with experiment is dif®cult because of lack of reliable vibrational spectroscopyresults on well-characterized, single-crystalline rutile. However, the calculations seem to overestimate the amount of this shift [523] with one exception [516]. The CO/TiO2(1 1 0) system is a good test case for a re®nement of computational techniques. Several authors explored systematically how these theoretical results are in¯uenced bythe embedding method of the clusters, the basis set, and various corrections [242,513,516±520]. U. Diebold / Surface Science Reports 48 (2003) 53±229 159

Table 11 Comparison of calculated equilibrium distances, CO vibrational shift Dn with respect to the free CO frequency, and adsorption a energies for CO molecule adsorbed on the TiO2(1 1 0) surface at different coverages, y Ê Ê À1 Reference Model/orientation y Ti±C or Ti±O (A) C±O (A) Dn (cm ) Eads (kcal/mol) Kobayashi and Yamaguchi [514] Cluster/±CO 2.500 1.150 17.00 Fahmi and Minot [515] Slab/±CO 1 2.379 1.123 75 10.87 Fahmi and Minot [515] Slab/±OC 1 2.340 1.132 139 6.73 Pacchioni et al. [517] Slab±cluster/±CO 2.328±2.381 1.155±1.126 80±140 16.14±18.44 Pacchioni et al. [517] Slab/±CO 1 11.5 Reinhard et al. [520] Slab/±CO 0.5 2.34±2.73 5.4±17.5 Reinhard et al. [520] Cluster/±CO 2:50 Æ 0:20 1.110 6±12 Casarin et al. [242,518,519] Cluster/±CO 2.344 1.126 56±72 6.7±26.5 Sorescu and Yates [513] Slab/±CO 1 2.376 1.121 64 7.97 Sorescu and Yates [513] Slab/±CO 0.5 2.320 1.119 11.10 Sorescu and Yates [513] Slab/±OC 1 2.701 1.128 64 0.62 Sorescu and Yates [513] Slab/±OC 0.5 2.666 1.129 2.71 Yang et al. [516] Slab/±CO 1 2.37±2.54 1.129±1.126 28 18.21±5.76 Yang et al. [516] Slab/±OC 1 2.45 6.68 Linsebigler et al. [510] TPD 0 9.9 Linsebigler et al. [510] TPD 0.68 7.7 a Notations ±OC and ±CO pertain to the two possible orientations of the CO molecule on the rutile (1 1 0) surface, i.e., with C and O atoms towards the surface. Adapted from [513].

Based on their TPD measurements, Linsebigler et al [510] suggested that a CO moietyadsorbed close to an oxygen vacancy experiences an additional interaction with the vacancy that leads to a stronger bonding, Fig. 54. Onlyone calculation found an increase in binding energyof CO at defect sites [514], in agreement with experiment [510]. However, this calculation was performed with a rather small, unrelaxed cluster [514].

5.1.4.2. CO2. Carbon dioxide interacts onlyweaklywith the clean rutile (1 1 0) surface [131,277]. Promotion of a TiO2(1 1 0) surface with half a monolayer Na surface leads to CO2 adsorption at room temperature, this is discussed in the context of Fig. 42. On the clean surface, CO2 was found to bind to oxygen vacancies slightly more strongly than to regular lattice sites, with desorption temperatures in TPD at 166 and 137 K, respectively [277]. The adsorption mechanism is precursor-mediated. From

HREELS data it was concluded that the molecule is linearlybound. The co-adsorption of CO 2 with H2O has been investigated byHenderson [277].CO2 adsorption is blocked bywater. When H 2O is dosed on a CO2-covered surface, the CO2 is displaced. The two molecules onlyinteract when dosed simultaneously, and possiblyform a bi-carbonate species [277]. Irradiation of ultraviolet light causes the reduction of

CO2 when H2O is present on TiO2(1 0 0) and TiO2(1 1 0) single crystal surfaces [524].

5.1.5. Nitrogen-containing molecules (N2, NO, NO2,N2O, NH3)

5.1.5.1. N2 (Table 12). The adsorption of N2 is important in the BET analysis of surface areas in powder materials. In addition, UV illumination of TiO2 can be used for the photo-oxidative fixation of molecular nitrogen as an NOx species [525]. 160 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 12

Surveyof the adsorption of N-containing inorganic molecules on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference

N2 Rutile (1 1 0) Cluster calculations, physisorption, end-on con®guration [244] exhibits attractive potential Rutile (1 1 0) Monte Carlo simulation, at low pressures: adsorption [526,527] on Ti in end-on orientation; at higher pressures: adsorption on bridging oxygens in side-on orientation

NO Rutile (1 1 0) TPD, NO reacts with O vacancies under formation of N2O [490] and oxidation of vacancy Rutile (1 1 0) TPD, DFT slab calculations, physisorbed NO for low coverages, [529] tilted molecule N-end down; lateral interaction for higher coverages;

N2O2 intermediate results in production of N2O Rutile (1 1 0) Cluster calculations, adsorption of NO on stoichiometric and [778] O-de®cient surface Rutile (1 1 0) Photodesorption, rapid depletion of adsorbed NO, production of [530]

N2O is dominant, production of NO after longer irradiation, see Fig. 55 NO Na Rutile (1 1 0) XPS, UPS, EELS, LEED, Na-dosed surface is activated for NO [305] adsorption

N2O Rutile (1 1 0) SHG, XPS, UPS, N2O heals defects, no N remains on the surface [501,532] NO2 Rutile (1 1 0) SXPS, EXAFS, DFT calculations, NO2 adsorbs predominantlyas [533] NO3 through disproportionation on Ti sites, subsurface vacancies are important

NH3 Rutile (0 0 1) UPS, TPD, adsorbs molecularlyat room temperature, desorbs at [536] 338 K, e-beam damage leads to NH2 and OlatticeH Rutile (1 1 0) UPS, adsorbs mainlymolecularly,desorption and dissociation at [537] higher temperatures Rutile (1 1 0) XPS, molecular adsorption, slightlylower coverage on surface [136,534] with vacancies; ESD, mainlydesorption on stoichiometric surface;

mainlydissociation on reduced surface; NH 3 overlayer suppresses O desorption from substrate Rutile (1 1 0) ESDIAD, upright molecule, H atoms rotate freely [535] Rutile (1 1 0) Auger-photoelectron coincidence spectroscopy, molecular adsorption; [540] vacancies are active sites for adsorption Rutile (1 1 0) STM [491] Rutile (1 1 0) Molecular dynamics simulations, upright molecule, orientation depends [539] on coverage

NH3 H2O Rutile (1 1 0) Periodic Hartree±Fock calculations, NH3 adsorbs molecularly; [538] co-adsorption with water leads to H bonds

NH3 O2 Rutile (1 1 0) TPD, O2 dissociation at point defects produces adatoms which [499] presumablyenhance dissociation of NH 3

The physisorption of N2 was treated theoretically [244,526,527]. According to cluster calculations the molecule physisorbs at TiO2 surfaces in an end-on con®guration [244]. The experimentallyfound high adsorption energyfor zero coverages was reproduced in these calculations. An interaction potential was constructed for N2/TiO2 clusters of various stoichiometries [527], and was used in Monte- Carlo calculations of N2 adsorption at 77 K [526]. For adsorption at low pressures (<1 Torr), every ®vefold coordinated Ti site is occupied byone N 2 molecule; lateral repulsion between molecules causes an arrangement of N2 in 1D zigzag chains. According to these calculations, the area covered byone N 2 U. Diebold / Surface Science Reports 48 (2003) 53±229 161 molecule (19.2 and 9.6 AÊ 2 in the monolayer regime and multilayer regime, respectively) is quite different from the one usuallyused for determination of surface areas (16.2 AÊ 2). Simulated adsorption isotherms also showed deviation from those expected from the BET model.

5.1.5.2. NO. Nitrogen oxides (NOx) are major contributors to acid rain and ground-level ozone pollution and TiO2 is used to photocatalytically oxidize NOx [528]. In field studies it was found that TiO2 under UV illumination converts nitric oxide (NO) to nitrogen dioxide (NO2) and HNO3. Molecular scale studies of NO adsorption on single-crystalline TiO2 surfaces are useful for gaining an understanding of the underlying mechanisms [304,529,530]. For low coverage, weak molecular adsorption was observed [529,530]. Na deposition activates the

TiO2(1 1 0) surfaces for adsorption of NO and the formation of nitride [304]. Desorption of NO from the clean surface occurs around 127 K [529,530]. The temperature of this NO TPD peak is almost independent of coverage [529,530]. Lateral interactions of the adsorbed NO molecules lead to the formation of an N2O reaction product. The production of this species is observed in TPD above certain critical coverages. N2O molecules desorb at 169 and 250 K when the surface was exposed to dosages higher than 5:5 Â 1014 and 2:2 Â 1015 molecules/cm2, respectively [529]. DFT slab calculations found as the favorable adsorption geometrya tilted NO molecule with the N end down

[529]. The calculations showed formation of an N2O2 species for higher coverages which is bound much stronger than NO. In a previous TPD study [490], it was found that NO reacts reductivelyat the 3 Ti sites at oxygen vacancies to produce N2O. The oxygen atoms in the adsorbate molecules are 3 extracted bythe TiO 2 surface and cause the oxidation of surface Ti sites [490]. DFT calculations [529] showed that other molecules, such as N2OorNO2 bind weaklyto the surface. The effect of UV irradiation on NO adsorbed at single-crystalline TiO2(1 1 0) was also investigated and the results were compared to experiments on compressed TiO2 powders [530,531]. Fig. 55 shows representative traces of photodesorption followed with a mass spectrometer. NO is depleted by3.6 eV photons with a quantum ef®ciencyof unity,and the dominant process for this depletion is formation of

N2O. After the initial rapid depletion of NO, a slow NO production is observed in Fig. 55. It was conjectured that the NO production is related to the capture of photoproduced N2O to the surface and its slow subsequent photodecomposition to NO(g). However, N2O adsorption and photodesorption experiments on TiO2 powder did not con®rm this hypothesis [530]. Evidence for photodesorption at photon energies below the 3.2 eV band gap has also been observed [531], probablydue to electronic excitations from defect-related ®lled electronic states in the band gap region.

5.1.5.3. N2O. The efficiencyto `heal' oxygenvacancies with N 2O was studied with photoemission spectroscopy [532] and second harmonic generation [501,532]. It was found that N2O fills oxygen vacancies at a similar rate as molecular oxygen. No nitrogen is deposited on the surface.

5.1.5.4. NO2. The main product of the adsorption of NO2 on TiO2(1 1 0) surface is surface nitrate, NO3, with a small amount of chemisorbed NO2 [533]. Photoemission data and DFT calculations suggest that this surface nitrate forms through a disproportionation process

2NO2;ads ! NO3;ads NOgas (6) after adsorption at Ti sites. Exposure of defect-rich TiO2 to NO2 at 300 K healed both, surface and subsurface defects. Because NO2 is a big molecule, this result implies that substrate O vacancies and 162 U. Diebold / Surface Science Reports 48 (2003) 53±229

15 Fig. 55. NO photodesorption from TiO2(1 1 0). In the upper panel is shown the decayrate of the m=e 31 signal with 15 15 time. The majorityof the signal originates from the NO cracking product of an N2O photoproduct (represented bythe continuous line in the upper panel). The cross-hatched region represents an additional photodesorption signal that originates as a result of the simultaneous production of NO which also yields an m=e 31 ion. From Rusu and Yates [531]. # 2000 The American Chemical Society.

related defects migrate towards the surface in the presence of NO2; a thesis that is supported byDFT calculations [533]. This is yet another example for the importance of subsurface defects for the surface chemistryof TiO 2. Such mechanisms are of potential importance when using an oxide for trapping or destroying NOx species in the prevention of environmental pollution [533]. U. Diebold / Surface Science Reports 48 (2003) 53±229 163

5.1.5.5. NH3. Ammonia adsorbs molecularlyon TiO 2(1 1 0) as well as TiO2(0 0 1) surfaces at room temperature. This was confirmed in several experiments [499,534±537] as well as theoretical calculations [538,539]. It binds with the N-end down to the fivefold coordinated Ti sites. ESDIAD measurements indicated that the H atoms are either randomlyoriented or that the molecule is freely rotating around its C3v axis [535]. These ESDIAD measurements were taken in a mass-resolved time-of- flight mode in order to suppress the O signal from the TiO2(1 1 0) substrate that is emitted in normal direction [135]. The NH3 overlayer efficiently neutralizes this O signal [136]. On the stoichiometric substrate, electron bombardment results predominantlyin desorption of the whole molecule and has a relativelysmall dissociation cross-section. On a sputter-reduced substrate, however, N±H bond scission dominates and e-beam bombardment can be used to form a Ti±O±N compound even at low temperatures

[534,535]. Surprisingly, the NH3 saturation coverage at room temperature is smaller on a UHV-annealed surface with point defects than on a stoichiometric surface. This can be explained if one assumes that NH3 molecules bind more stronglyto defect sites and effectively`block' neighboring, regular Ti(5) adsorption sites [534,540]. When an oxygen deficient TiO2(1 1 0) surface is exposed NH3, the vacancies are filled up in a dissociative process. The resulting adatom leads to increased dissociation of the NH3 molecule, similar to the case of O2/H2O co-adsorption [499].

5.1.6. Sulfur-containing molecules (SO2,H2S,Sn) The removal of sulfur compounds from feedstocks is an important issue in the petrochemical industry. Sulfur poisoning of catalysts is a serious and costly problem in many re®ning processes. Reduction of sulfur content reduces corrosion, and environmental regulations limit the output of sulfur to the environment. Thus, fundamental insight into the interaction of S-containing molecules with model catalysts can possibly help to develop sulfur-resistant catalysts as well as catalysts for hydrodesulfurization. Technologically the Claus process

2H2S SO2 ! 3S #2H2O (7) is used for hydrodesulfurization. It is mainly carried out on alumina catalysts, but titania is also used [11]. Adsorption and reaction of all molecules involved in the Claus process have been investigated on

TiO2(1 1 0) surfaces (Table 13).

5.1.6.1. SO2

5.1.6.1.1. TiO2(1 1 0). Several groups have studied the adsorption of SO2 on TiO2(1 1 0), and somewhat conflicting reports were given on the surface chemistryof this system [541±545]. Smith et al. [543,544] 6 detected no change in photoemission spectra from TiO2(1 1 0) for SO2 exposures up to 10 L. It was concluded that SO2 interacts onlyweaklywith stoichiometric TiO 2(1 1 0). However, a `violent' interaction with sputtered surfaces was found, where SO2 dissociates at defect sites. In fact, these authors reported that SO2 exposure oxidizes sputter-reduced TiO2 surfaces completelywithin the information depth of photoemission. SO2-induced oxidation was also found on single-crystalline Ti2O3 surfaces [543,544]. This implies that sub-surface defects migrate to the surface where they react, similar to the process that was invoked in a more recent studyon NO 2 adsorption [533]. Smith and Henrich [544] found no indication for S±O bonds in XPS after exposure of SO2 to a sputtered surface, and concluded that the S atom binds to Ti sites, forming TiS2-like structures. 164 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 13

Surveyof the adsorption of S-containing inorganic molecules on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference

SO2 Rutile (1 1 0) UPS, XPS, weak interaction with stoichiometric surface, oxidizes [543,544] sputtered TiO2(1 1 0) completelyat room temperature 2À Rutile (1 1 0), XPS, UPS, LEED, SO3 on TiO2(1 1 0) and stepped [545] 3 2À (4 4 1) TiO2(4 4 1), reacts with Ti to form S Rutile (1 1 0) NEXAFS [541,542] Rutile (1 1 0) ESD, ESDIAD [546±549]

Rutile (1 0 0), NEXAFS, chemisorbed SO2 (with molecular plane parallel to the [550,551] 2À 2À (1 Â 1) and (1 Â 3) surface) and SO4 ;SO2 reacts to SO4 at elevated temperature H2S Rutile (1 1 0) Photoemission on Ar bombarded surface initial dissociation [544,552] followed bymolecular adsorption Rutile (1 1 0) Cluster calculation, weak adsorption [242,519] Anatase (1 0 1) First-principles MD calculations, molecular adsorption [553]

Sn (n 2) Rutile (1 1 0) STM, XPS, LEED, photoemission, adsorbs at Ti(5) and O [71,77,555,556,559] vacancies at room temperature, replaces O at high temperature (Fig. 56); (3 Â 1), (3 Â 3), and (4 Â 1) superstructures at 100±400 8C, depending on coverage; replacement of O atoms is mediated bybulk defects, see Fig. 58 Rutile (1 1 0) Photoemission, DFT calculations, adsorption at Ti sites, O [557,558,779]

vacancies, and SOx at room temperature, S2 and Sn at low temperature; adsorption at high temperature involves bulk O vacancies

2À Onishi et al. [545] reported that SO2 adsorbs on TiO2 surfaces to form an SO3 -like complex. On a 2À 3 (stepped) TiO2(1 1 0) surface, SO3 was found as well as oxidation of Ti sites, which was probably connected with the dissociation of the molecule [545]. NEXAFS measurements in Thornton and 2À coworkers [541,542] point toward a chemisorbed SO2 species at 100 K, which reacts to form SO4 when the substrate is heated to temperatures 135 K. This surface sulfate species stays at the surface up to 450 K. A SO3-like species was identi®ed as the intermediate in this reaction. A model was suggested where the SO2 molecule is incorporated into the bridging oxygen rows [541]. A small amount of S2À species was also identi®ed on the sample and attributed to dissociation at surface vacancies. Electron-stimulated desorption of SO2 on TiO2(1 1 0) was investigated byde Segovia and co-workers [546±549]. Adsorption of SO2 as well as annealing an SO2-covered surface changes the O ion yield and energy distribution. The ESDIAD pattern was not changed from the one of a clean surface, however.

5.1.6.1.2. TiO2(1 0 0). Thornton's group also performed NEXAFS studies on SO2 adsorbed on TiO2(1 0 0) [550,551]. The surface chemistryon TiO 2(1 0 0) is also similar to the one on TiO2(1 1 0), and the results from the TiO2(1 0 0)-(1 Â 1) resembled those from (1 Â 3) reconstructed surfaces. SO2 chemisorbs on TiO2(1 0 0), and a 0.5 ML coverage was achieved at 110 K. Polarization- dependent NEXAFS measurements were performed at substrate temperatures of 130, 200, and 500 K, in 2À 2À order to determine the geometryof SO 2,SO3 , and SO4 , respectively. Models were proposed where 2À 2À SO2 adsorbs at exposed fivefold coordinated Ti sites, and both, SO3 and SO4 , form through interaction with the bridging oxygen rows [550]. U. Diebold / Surface Science Reports 48 (2003) 53±229 165

5.1.6.2. H2S. The interaction of H2S with TiO2 surfaces was studied bySmith and Henrich [544,552] using photoemission techniques. The molecule interacts weaklywith TiO 2. UPS spectra were interpreted as dissociation of the molecule for small exposures, followed bymolecular adsorption. A cluster calculation was performed with onlyone test geometry,i.e., with H 2S adsorbed with the S-end down and the hydrogens confined to the (0 0 1) plane [242,519]. A relativelylarge Ti±S distance (3.1 AÊ ) and a small interaction energy(7.0 kcal/mol) was found. As pointed out bythe authors of this study [242,519], these values should be treated with caution because of the limited test geometries considered and the neglect of interadsorbate interactions. DFT-based slab calculations [553] on anatase (1 0 1) predict molecular adsorption with the plane of the molecule parallel, due to interaction between H and O atoms at the ridges of the anatase (1 0 1) surface (see Fig. 28). An adsorption energyof 0.49 eV (11.3 kcal/mol) was calculated.

5.1.6.3. Elemental sulfur (Sn,n 2). The adsorption and reaction of elemental sulfur with TiO2 was studied bythis author's group as well as byHrbek, Rodriguez, and co-workers, see Table 13. Sulfur was dosed from a UHV-compatible, electrochemical source as described byBechtold and co-workers [554].

Such a source typically delivers elemental sulfur in the form of Sn n 2. The adsorption behavior, mechanism, and the resulting surface structures on TiO2(1 1 0) depend drasticallyon the sample temperature, see Fig. 56. When sulfur is dosed at room temperature, it dissociates and adsorbs at the ®vefold coordinated Ti atoms, probablyin an on-top con®guration [77,555,556]. This can clearlybe seen in Fig. 56A where S atoms are visible as large, white spots situated along the bright rows of the empty-states STM image. The S atoms are quite mobile in STM and no ordered superstructure appears at higher coverages. Sulfur also adsorbs readilyon oxygenvacancies as evidenced bythe disappearance of the defect-related band gap state in valence band photoemission spectra [556]. High-resolution core-level spectra were ®tted with three peaks that were attributed to SOx groups as well as sulfur bonded to Ti rows and O vacancies [557]. In addition, S2 and Sn species were detected on the surface after sulfur was dosed at low temperature. When S is dosed at elevated temperature (120 8C), or when a surface is ®rst exposed to S at room temperature and then heated up, S switches adsorption sites from the position of the Ti rows to the position of the bridging oxygen rows (Fig. 56B). The S2p core level binding energyin XPS shifts to lower binding energy, contrary to what is expected for SOx formation. Also, the Ti2p core levels show a

Fig. 56. Empty-states STM images of sulfur adsorbed on TiO2(1 1 0). (A) At room temperature, bright S atoms are located at the bright rows (at the ®vefold coordinated Ti atoms). (B) Sulfur dosed at 300 8C. Sulfur is located at the dark substrate rows (at the position of the bridging oxygen rows). 166 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 57. STM and LEED of 0.55 ML sulfur adsorbed at 400 8C, forming a 3 Â 3-S superstructure. In the STM image (300 Ð Â 220 Ð, 1.4 V, 0.5 nA) bright double rows run along the [0 0 1] direction. Occasionally, additional S atoms adsorb within these rows, some of which are marked byarrows. A model was suggested where these rows represent sulfur atoms that have replaced in-plane oxygen atoms. From Hebenstreit et al. [555]. # 2001 Elsevier. distinct shoulder [77,220,557], and a Ti3d-derived state appears in the band gap [556]. This points towards a replacement of bridging oxygen atoms, rather than an adsorption on them. Depending on the sample temperature and S coverage, different superstructures evolve at the TiO2(1 1 0) surface [77,555].A 3 Â 1-S, 3 Â 3-S, and 4 Â 1-S structure was observed with LEED and STM. An example for a S-induced superstructure is shown in Fig. 57. The preparation parameters, structural units, and adsorption models of these structures are discussed in [77,555]. Theyconsist of S atoms replacing surface oxygen atoms, both the in-plane and bridging oxygens. The electronic structure of these high-temperature adsorption states was mapped out with resonant photoemission [556]. Hrbek et al. [558] observed that virtuallyall surface oxygencan be replaced byS within the information depth of XPS. The replacement of oxygen by S is thermodynamically `uphill', i.e., the heat of formation of Ti±S bonds is lower than that of Ti±O bonds. Also, the reaction takes place at rather low temperature (120 8C), where desorption of O atoms is unlikely. There is experimental and theoretical evidence [71,557±560] that bulk defects are involved in the surface sul®dation process. For example, it was observed that S, adsorbed at room temperature, does not react with the surface but onlydesorbs on a light, slightlyreduced sample. However, when the experiment was repeated with a darker, more bulk- reduced sample, some of the S stayed at the surface up to fairly high coverages, and this S switched the adsorption site. Fig. 58 shows another example of the in¯uence of bulk defects on S adsorption [559].A pristine, stoichiometric sample was cleaned and then exposed to S at 300 8C; at this temperature, S replaces oxygen. The S coverage was measured with XPS. The sample was then cleaned and heated for several hours in UHV to a temperature of 700 8C. This caused an increase in the level of bulk reduction, and a darkening of the sample color. The adsorption experiment was then repeated with the same S ¯ux and exposure time. The amount of adsorbed S increased with the level of bulk reduction. The following model was suggested to account for the observed behavior: at elevated temperatures, bulk defects (O vacancies and Ti interstitials) become mobile and can migrate to the surface. Sulfur adsorbs at the U. Diebold / Surface Science Reports 48 (2003) 53±229 167

Fig. 58. Sulfur coverage of S/TiO2(1 1 0) after adsorption at 300 8C. Each data point represents a separate adsorption experiment with similar S ¯ux and exposure time. The cleaned sample was reduced between the experiments byheating at 700 8C in UHV. The more reduced the bulk of the sample becomes the more S adsorbs at the surface. The second increase of the lower curve is caused bythe depletion of Mg-segregated impurities due to cleaning cycles (sputtering and annealing). From Hebenstreit et al. [559]. # 2001 Elsevier.

®vefold coordinated Ti atoms on the TiO2(1 1 0) surface. It is weaklybound, rather mobile, diffuses along the Ti rows, and desorbs after a while. If the S atom encounters an oxygen vacancy (or a Ti interstitial), it reacts, binds more strongly, and becomes trapped. The probability of the latter process increases with the concentration of bulk defects, and the ¯ux of defects to the surface. It was pointed out that such a bulk-defect dependent adsorption process might be important in the behavior of titania as desulfurization catalyst and in the design of more S-resistant catalytic materials.

5.1.7. Halogen-containing molecules (Cl2, CrO2Cl2, HI)

5.1.7.1. Cl2. The adsorption of Cl at TiO2(1 1 0) was studied experimentally [106,109,561] as well as theoretically [106,562]. The chlorine was dosed from an electrochemical source [554] in the form of Cl2. It dissociates at room temperature and adsorbs at the fivefold coordinated Ti atoms in an on-top configuration [106,562]. Oxygen vacancies, if present, are filled preferentially [106,561], Table 14.

Table 14

Surveyof the adsorption of halogen-containing inorganic molecules on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference

Cl2 Rutile (1 1 0) STM, dissociates, high transient mobility(`cannon-ball' [109] adsorption) at low coverages, see Figs. 59 and 60 Rutile (1 1 0) DFT calculations, STM, photoemission, precursor adsorbs [106] at Ti4 ions, adsorption in oxygen vacancies is favored Rutile (1 1 0) DFT calculations [562] Rutile (1 1 0) STM, XPS, photoemission, adsorbs at Ti(5) at room [561] temperature, replaces O at elevated temperatures

CrO2Cl2 Rutile (1 1 0) TPD, AES, workfunction, SSIMS, XPS [570] HI Anatase (1 0 1) First-principles MD calculations, dissociative adsorption [553] 168 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 59. Left: STM image of a TiO2(1 1 0) surface exposed to 0.07 Cl2 at room temperature (400 Ð Â 380 Ð, Vs 1:6V, It 1:6 nA). Adsorbed chlorine atoms appear as bright round spots on the bright rows parallel to the [0 0 1] direction (on ®vefold coordinated Ti atoms). Most Cl atoms are paired. Some of the Cl±Cl pairs and single chlorine adatoms are marked with circles and squares, respectively. Right: autocorrelation of Cl±Cl distances in STM images with the same Cl coverage as shown on the left side. Probabilities are given for areas of three unit cells to smooth the statistical scatter and to account for the limited accuracyof position determination along [0 0 1]. Probabilities that are signi®cantlyhigher than the average value of 2% are printed in bold. From [109].

The adsorption mechanism of Cl is quite interesting, as it represents an experimental veri®cation of `hot' surface atoms (atoms with high kinetic energythat result from a dissociative adsorption process) which is a matter of some debate in the literature [563±567]. At low coverages, deposited at room temperature, clearlycorrelated Cl±Cl pairs are observed [109], see Fig. 59. This was explained with the

Fig. 60. Model for the high transient mobilityof chlorine atoms that gives rise to the widelyseparated Cl±Cl pairs in Fig. 59. Labels (A)±(E) are explained in the text. From [109]. U. Diebold / Surface Science Reports 48 (2003) 53±229 169 model in Fig. 60. The Cl2 molecule approaches the surface (A). Cl2 dissociation on Ti atoms is an exothermic process, and part of this excess energycan be transferred into kinetic energy.If a Cl 2 molecule dissociates in a trough (B) the Cl atoms staycon®ned to the Ti rows and can possiblymove a long distance before their energyis accommodated (B). The distribution of Cl pairs in Fig. 59 points towards a correlation across bridging oxygen rows as well, however. Such widely spaced

Cl±Cl pairs could be explained bya `cannon-ball' mechanism where the Cl 2 molecule adsorbs in a approximatelyupright position (E). One Cl atom `shoots' out into the vacuum, similar to an abstractive adsorption process observed for halogens on Si [568,569]. Because of the high electronegativityof chlorine, this atom will likelybe negativelycharged, and attracted byits own image charge. It was suggested [109] that, depending on the orientation of this nanoscopic `cannon', this Cl atom mayland in the neighboring trough or even further awayfrom the original adsorption point. (The con®gurations labeled D and C in Fig. 60 describe the (unlikely) scenario that Cl2 dissociates in a precursor that is oriented perpendicular to the Ti rows (D) and the van-der Waals radius of an ClÀ ion, respectively.) When the `cannon' is oriented perpendicular to the surface, the Cl is emitted into the vacuum. The presence and number of `single' Cl atoms in Fig. 59 was taken as an indicator for this `cannonball' process. The described mechanism was based on STM images alone. More recent photoemission results have shown that Cl adsorbs also in O vacancies, as the defect-related gap state in photoemission spectra is quenched rapidlywhen Cl is dosed to a slightlydefective surface at room temperature. FLAPW calculations indicate that Cl atoms that are adsorbed in oxygen vacancies will not be observed with STM, as theyare expected to have an image contrast similar to bridging oxygenatoms [106]. This implicates that the `single' Cl atoms in Fig. 59 could be the `leftover' partner from a dissociated Cl2 pair where the other Cl atom ®lls a vacancyin the bridging oxygenrows. As pointed out in [561] these recent insights would then lead to a slight modi®cation of the originallyproposed model. The Cl molecule could also `explode' above the surface. This explosion would be initiated bycapture of an À À electron from the surface which would result in an unstable Cl2 moiety. The resultant Cl ions could possiblycover large distances above the surface before theyland in a trough. The FLAPW calculations mentioned above also predicted that adsorption of a Cl atom in an O vacancyis an energeticallymuch more stable situation as compared to Cl adsorption at the sites of the ®vefold coordinated Ti atoms [106]. STM measurements of chlorine adsorption [561], performed at elevated temperature, con®rm this prediction. When Cl is dosed above a temperature of 150 8C, the Cl replaces bridging oxygen atoms. Other features, possibly some sort of Ti±O±Cl complex, are also formed. The high-temperature adsorption process is mediated bybulk defects, similar to the process discussed in the context of S adsorption (Fig. 58). The calculations also investigated the STM image contrast of adsorbed Cl [106]. As pointed out several times, the STM image contrast on TiO2 is largely electronic. Hence it is somewhat surprising that a negativelycharged atom with hardlyanyemptyDOS should be imaged bright in empty-states STM images. Analysis of calculated charge-density contours shows that the un®lled Ti states underneath the adsorbed Cl atom are considerablychanged which leads to the observed bright spots in STM.

5.1.7.2. Other halogen-containing molecules. Hazardous hexavalent Cr(VI) is often present in waste water streams, and TiO2 can be useful for the photoreduction of Cr(VI) to the more benign trivalent Cr(III). With this application in mind, the adsorption of chromyl chloride on TiO2(1 1 0) was studied with surface science techniques byAlam et al. [570]. It was suggested that the reduction and potential 170 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 15

Surveyof the adsorption of rare gas atoms on TiO 2 surfaces Substrate Techniques/adsorption/reaction Reference

Ar H2O Rutile (1 0 0), (1 1 0) Computer simulation of adsorption isotherms [572] Xe Rutile (1 1 0) Monte Carlo simulation [573] Rutile (1 1 0) Simulations, Xe behaves fluid-like in 1D troughs along [0 0 1] [574]

immobilization of Cr(VI) species on TiO2 materials mayoccur thermallyif the appropriate surface defect sites are present. The adsorption and photo-oxidation/desorption of methyl halides is treated in Section 5.3.1. The adsorption of hydrogen iodide on anatase (1 0 1) was treated with first-principles molecular dynamics simulations by Selloni et al. [553]. The molecule dissociates and binds with an adsorption energyof 0.67 eV (15.45 kcal/mol).

5.1.8. Rare gases (Ar, Xe)

This author is not aware of anyexperimental studies of rare gas adsorption on single-crystalline TiO 2 surfaces. One exception is a studyof Pt growth on TiO 2, where photoemission of adsorbed Xe (PAX) was used as an spectroscopic technique to measure changes in the local work function [571]. The adsorption of Ar [572] and Xe [573,574] has been simulated with Monte-Carlo methods, however. It should be mentioned here that substantial amounts of Ar can be found in AES and XPS measurements of previouslysputtered samples, even if their surfaces have been fullyannealed. Ar atoms that are trapped sub-surface have an in¯uence on surface chemistryon metals [575,576], but it has not been investigated if a similar effect exists on TiO2 (Table 15).

5.2. Adsorption and reaction of organic molecules

This part is divided into subsections where organic adsorbates are looselygrouped bytheir functional groups. The main results of adsorption and reaction studies on these adsorbates are summarized in table format. Formic acid and other carboxylic acids have become the most investigated organic molecules on single-crystalline TiO2 surfaces (Table 16(a)). Because manyof the higher carboxylicacids follow the behavior of formic acid, the adsorption and reaction of HCOOH is discussed in some detail in the ®rst part of Section 5.2.1. A large number of reactions of higher organic molecules was studied in Barteau's group. Differently prepared TiO2(0 0 1) surfaces were used, in particular two differentlyreconstructed surfaces, which are commonlyreferred to as `{1 1 4}'- and `{0 1 1}-faceted' surfaces, 3 see Section 2.4. The UHV studies were often compared to results from powder materials. A particularlyintriguing reaction is the isomerization of alkynes (Section 5.2.4). This reaction is verysensitive to the reduction state of the sample. Some of the results are summarized in the following, but the reader is also referred to a series of excellent review articles byBarteau et al. [6,551,577±580] on the adsorption and reaction of higher organic molecules.

3 According to recent STM results, the assignment of these reconstructions as faceted surfaces is questionable, see Section 2.4 and Fig. 25. An atomic model is still lacking at this point, thus these surfaces are still quoted as `{1 1 4}-faceted' and `{0 1 1}-faceted'. U. Diebold / Surface Science Reports 48 (2003) 53±229 171

Table 16

Surveyof the adsorption and reaction of organic molecules on TiO 2 surfaces Molecule Substrate Adsorption/reaction Reference

(a) Surveyof the adsorption and reaction of carboxylicacids on TiO 2 surfaces Formic acid, All rutile surfaces Dissociative adsorption as formate plus hydroxyl; Reviews, HCOOH molecular at low temperatures and higher coverages see [551] Rutile (1 1 0) Orders as p(2 Â 1) structure with 0.5 ML coverage [581,599] above 250 K; STM shows c(4 Â 2) overlayer at 22% of the saturation coverage Rutile (1 1 0) Adsorption geometryfrom photoelectron [585,586] diffraction, see Fig. 61 Rutile (1 1 0) Adsorption geometryfrom NEXAFS [589] Rutile (1 1 0) Vibrational spectra, detection of minority [588] species (Fig. 61) from IRAS Rutile (1 1 0) Vibrational spectra from HREELS [582,585] Rutile (1 1 0) Non-contact AFM of single formate ions [594,595,600] Rutile (1 1 0) STM of single formates [108] Rutile (1 1 0) STM: diffusion mainlyalong [0 0 1] direction, [596,598] see Fig. 62 Rutile (1 1 0) Theory: diffusion in¯uenced by neighboring H atoms [591] Rutile (1 1 0) Theory: (CRYSTAL program) acidic cleavage; [592] leads to bi-dentate formate Rutile (1 1 0) Theory: DFT and pseudo-plane wave, [102] dissociative adsorption Rutile (1 1 0) switchover from dehydration [581,606] (DCOOD ! CO2 H2O) to dehydrogenation (DCOOD ! CO and H2) around 500 K Rutile (1 1 0) CO main decomposition product, small amounts [582]

of H2CO, desorption of H2O before onset of decomposition; scrambling with 18O from 18O-enriched surface Rutile (1 1 0)- STM: no adsorption at (1 Â 2) rows [122] (1 Â 2) (added

`Ti2O3'row) Highlyreduced STM: adsorption at cross-links, high-temperature [584] rutile (1 1 0), reaction, produces (1 Â 1) islands (Fig. 63) cross-linked (1 Â 2) Rutile (1 1 0), Dehydration and dehydrogenation independent of [780] Ar-bombarded annealing historyof ion-bombarded sample Defective rutile (1 1 0) Coverage scales with reduced Ti sites [593] (defects bysputtering, e-beam)

Thin TiO2 ®lm Adsorption as bi-dentate on (1 1 0)-terminated [432] surface, as monodentate at c(2 Â 6) surface Rutile (4 4 1) Adsorption at step edges in possiblytilted [545] con®guration Rutile (1 0 0)-(1 Â 3) Scrambling with 18O from 18O-enriched [196]

surface, decomposition at 550 K to CO, H2CO; trace of HC=CH; H2O desorbs before HCOOH decomposition 172 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference Rutile (1 0 0)-(1 Â 1) Main decomposition product CO, no HCBCH or [196]

H2CO; H2O desorbs before decomposition Rutile (0 0 1)- Unimolecular HCOOH decomposition above [603]

`{1 1 0} faceted' 550 K to CO and CO2 Rutile (0 0 1)- Formation of H2CO in addition to CO and CO2 [603] `{1 1 4}-faceted'

Rutile (0 0 1) On reduced surface: H2CO via unimolecular [583] decomposition, oxidizes surface; on oxidized

surface: H2CO via bimolecular coupling of two HCOOÀ molecules Anatase (1 0 1) Theory: DFT slab calculations, HCOOH [605] adsorbs molecularlyas monodentate, desorption promoted byco-adsorbed water, NaCOOH adsorbs dissociativelyas bi-dentate Acetic acid, All rutile surfaces Dissociative adsorption to acetate and hydroxyl [551]

CH3COOH Rutile (1 1 0) STM, ESDIAD, LEED: (2 Â 1) overlayer, [608,614, molecule stands upright 781,782] Rutile (1 1 0) NEXAFS: molecule stands upright [589] Rutile (1 1 0) Non-contact AFM measurements [600,610,611] Rutile (1 1 0) STM at 580 K shows decrease of acetate [609] coverage, possiblyvia decomposition into ketene Rutile (1 1 0)- No ordered overlayer in LEED, saturation [608,781] (1 Â 2) acetate coverage at room temperature ca. half of

(added Ti2O3 row) that on the TiO2(1 1 0) Rutile (0 0 1) Adsorbs both molecularlyand dissociativelyat [604] sputtered; `{0 1 1}- 200 K, onlydissociativelyat 300 K; decomposition faceted'; `{1 1 4}- products depend on surface pretreatment, see Fig. 66 faceted' Propanoic acid, Rutile (1 1 0) NEXAFS: p(2 Â 1) overlayer at 300 K, acetate [551,589]

C2H5COOH in a bi-dentate coordination and upright position Rutile (0 0 1) TPD; dominant reaction products: CO, CO2, [604] sputtered; `{0 1 1}- C4 products (butene, butadiene), acrolein; divinyl faceted'; `{1 1 4}- ketone (onlyon `{1 1 4}-faceted' surface) faceted' Acrylic acid, Rutile (0 0 1) TPD, XPS, scanning kinetic spectroscopy; reaction [612]

CH2=CHCOOH sputtered products contain ethene, ethyne, butene, butadiene, divinyl ketone (only on `{1 1 4}-faceted' surface) Benzoic acid, Rutile (0 0 1) Adsorbs both molecularlyand dissociativelyat [604]

C5H2COOH sputtered; 200 K, onlydissociativelyat 300 K; decomposition `{0 1 1}-faceted'; products depend on surface pretreatment `{1 1 4}-faceted' Rutile (1 1 0) ESDIAD and LEED: p(2 Â 1) overlayer at 300 K, [551,613] benzoate with upright aromatic ring; dimerization in STM U. Diebold / Surface Science Reports 48 (2003) 53±229 173

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference Bi-isonicotinic acid Rutile (1 1 0) NEXAFS, XPS, Hartree±Fock calculations; see [615,783] Fig. 67 for adsorption geometry; molecular at multilayers

2À Oxalic acid, Rutile and anatase Dissociative adsorption as oxalate (C2O4 ) ions, [617] HOOC±COOH `polymers' more stronglyon `anatase' than `rutile' models Glycine Rutile (1 1 0) UPS, SXPS: adsorbs as zwitterionic ion; high [618] cross-section for photon damage

Rutile (1 1 0)-(1 Â 2) UPS, SXPS [619] Maleic anhydride Rutile (0 0 1), TPD, semi-empirical calculations show dissociation [620] sputtered of one COC bond upon adsorption; decomposition into

CO, CO2, acetylene (HCBCH), ketene (H2C=C=O) vinylacetylene (HCBCCH=CH2), butene, traces of butadiene (H2C=CHCH=CH2), and benzene (C6H6)

(b) Surveyof the adsorption and reaction of alcohols on TiO 2 surfaces Methanol, Various TiO2 For a review, see Ref. [551, Chapter 10] CH3OH surfaces Rutile (1 1 0) XPS, UPS, LEED, TPD: molecular adsorption [545] at 289 K, no ordered overlayer observed Rutile (1 1 0) TPD, HREELS, SSIMS: majorityof adlayeris [621] molecular, some dissociative adsorption, see Fig. 68,

recombines upon annealing; CH3OH the only desorption product; co-adsorbed water has little or no in¯uence on surface chemistry;

co-adsorption of O2: O adatoms, produced by adsorption at vacancies at 300 K lead to increased dissociation; molecularlyadsorbed

O2 at 150 K oxidizes CH3OH to H2CO Rutile (1 1 0) Theory: dissociation and molecular adsorption [102,623] have equivalent energyin high-coverage limit, dissociation preferred at low coverage Rutile (1 1 0) Electron-bombardment of methanol adsorbed at 135 K [151] Rutile (1 1 0) Theory: electronic structure calculations, [241] molecular adsorption is thermodynamically favored, but energetic barrier for proton transfer Rutile (1 0 0) Theory: electronic structure calculations, molecular [241] adsorption is thermodynamically favored Rutile (4 4 1) Molecular adsorption [545] Rutile (0 0 1) Ar Molecularlyadsorbed methanol desorbs below 300 K; [627] bombarded, 50% dissociated molecules recombine at 365 K,

`{0 1 1}-faceted', remaining methoxides decompose to CH4, dimethyl `{1 4 4}-faceted' ether (CH3±O±CH3), formaldehyde (CH2O) and CO; selectivityof reaction products depends on surface 174 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference Rutile (0 0 1) with UPS, TPD: dissociative adsorption at 300 K, [628] point defects coverage-dependent dissociation energy, main

thermal desorption products H2, CO, and CH3OH; minor quantities of CH4,CH2O, and H2O Ethanol, C2H5OH Rutile (0 0 1), TPD: both molecularlyand dissociativelyadsorbed [629] `{0 1 1}-faceted' at 200 K, onlydissociativelyat 300 K. Desorption

products are ethylene (C2H4) and acetaldehyde (CH3CHO) D-Ethanol (EtOD) Rutile (1 1 0) TPD: formation of surface ethoxygroups by [630] and TEOS dissociative adsorption of deuterated ethanol or TEOS, co-adsorbed with water and hydroxyls; two ethoxyspecies: one that is bound to a Ti(5) atom, desorbs as ethanol at 250±400 K; second ethoxy species adsorbs at oxygen vacancies and decomposes into ethylene and ethanol at 650 K n-Propanol Rutile (0 0 1), TPD: dissociative adsorption; predominantly [629]

(CH3±CH2±CH2±OH), `{0 1 1}- recombinative desorption; n-propanol 2-propanol faceted' decomposition into C3H6 and C2H5CO; 2-propanol decomposition into C3H5

2-Propanol Rutile (1 1 0), Rutile (1 1 0): photocatalytic reaction to acetone [666±668]

(1 0 0) and water in the presence of O2; rutile (1 0 0): thermallyactivated dissociation

(c) Surveyof the adsorption and reaction of aldehydeson TiO 2 surfaces Formaldehyde, Rutile (1 1 0) TPD: deoxygenation at surface Ti3 sites when [490]

H2CO point defects are present Rutile (0 0 1) TPD: oxidized surface: formation of methanol via [632] the Cannizzaro reaction; on reduced surface formation of methanol, CO and CO, probablythrough

complete decomposition into Cads,Hads, and Olattice Acetaldehyde, Rutile (0 0 1) TPD: aldol condensation to form crotonaldehyde, [633,634]

H3C±C=O CH3CH=CHCHO, and crotyl alcohol (CH3CH=CHCH2OH) on stoichiometric surfaces; butene, H3C±CH2±CH2±CH3 on reduced surfaces Benzaldehyde Rutile (0 0 1) TPD: on reduced surfaces, reductive coupling to [635,636]

stilbene , NEXAFS

investigations of reaction intermediates Acetone, Rutile (0 0 1) TPD: on reduced surfaces major pathwaythe [638] reductive coupling to symmetric ole®ns with twice the carbon number of the reactant 2 acetone ! 2,3-dimethyl-2-butene U. Diebold / Surface Science Reports 48 (2003) 53±229 175

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference Acetophenone Rutile (0 0 1) TPD: on reduced surfaces major pathwaythe [638] reductive coupling to symmetric ole®ns with twice the carbon number of the reactant 2 acetophenone ! 2; 3-diphenyl-2-butene

, minor

reaction path to pinacol

p-Benzoquinone Rutile (0 0 1) TPD: reduction and coupling reactions to benzene, [639,640]

biphenyl terphenyl, and phenol

Cyclohexanone Rutile (0 0 1) TPD: on sputter-reduced surface reduction to [640]

cyclohexanol and reductive

coupling to H10C6=C6H10 Cyclohexenone Rutile (0 0 1) TPD: on sputter-reduced surface reduction to [640]

cyclohexenol and reductive

coupling to H8C6=C6H8

(d) Surveyof adsorption and reactions related to the trimerization of alkyneson TiO 2 surfaces Acetylene, HCBCH Reduced rutile (0 0 1) TPD: predominantlycyclomerizationto [641]

benzene

Methylacetylene Reduced rutile (0 0 1) TPD: cyclotrimerization to trimethylbenzene [641,642] (butyne),

CH3±CBCH 176 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference 2-Butyne Reduced rutile TPD: predominantlycyclomerization to [641] (di-methylacetylene) (0 0 1)

CH3±CBC±CH3

hexamethylbenzene

tert-Butylacetylene Reduced rutile TPD: cyclization to the C18-cyclotrimer [643] (0 0 1)

tri-tertbutylbenzene (Me stands

for CH3)

Trimethylsilyl-acetylene, Reduced rutile TPD: cyclotrimerization to tri-trimethylsilylbenzene [644]

(CH3)3Si±CBCH (0 0 1)

on reduced titania is possible;

but principal product is trimethylvinylsilane,

(CH3)3SiCH=CH2 Allene, H2C=C=CH2 Rutile (0 0 1) TPD: reduced surfaces: principal reaction [645] hydrogenation to propylene, H3C±CH=CH2; minor channels: dimerization to dimethylene,

cyclobutane , benzene, and an open-chain

C6H10 dimer; dimerization is suppressed on oxidized surfaces Cyclooctatetraene Rutile (0 0 1) TPD: converts to benzene and cyclooctatriene [646,647]

on reduced TiO2, not a major intermediate of alkyne cyclooligomerization; NEXAFS of adsorption geometry

(e) Surveyof adsorption and reactions of pyridine,its derivates, and other aromatic molecules on TiO 2 surfaces Pyridine Rutile (1 1 0) TPD, XPS, STM, MD calculations: pyridine is [650] weaklyphysisorbed,not bound to speci®c sites

at TiO2(1 1 0) terraces; TPD: two peaks at 220 and 270 K, multilayers at 160 K; XPS: no peak shift of N1s, STM: high mobility U. Diebold / Surface Science Reports 48 (2003) 53±229 177

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference Rutile (1 1 0) STM: diffuses at room temperature, strongly [124] adsorbed at fourfold coordinated Ti atoms at step sites, activityof Ti step atoms depends on step orientation, exchange between pyridine adsorbed at step sites and terrace sites Rutile (1 1 0) STM at elevated temperature and pyridine [652] background pressure: condensation at partially hydrogenated pyridines Rutile (1 1 0) Theory: DFT calculations of binding site [653] 4-Methylpyridine Rutile (1 1 0) STM: three adsorption geometries identi®ed [654]

2,6-Methylpyridine Rutile (1 1 0) TPD similar to pyridine; STM: molecules are [650] somewhat less mobile than pyridine

m-Xylene Rutile (1 1 0) TPD similar to pyridine [650]

Benzene Rutile (1 1 0) TPD: two peaks at 200 and 260 K, no multilayers [650] Rutile (1 0 0)- AES, UPS, XPS: adsorbs tilted on (1 Â 3), ¯at [651] (1 Â 1) and (1 Â 3) on (1 Â 1) at submonolayers; electron-stimulated processes: X-rayor electron irradiation induces polymerization

Phenol Rutile (1 1 0) STM in air [648,649]

(f) Surveyof adsorption and reactions of silanes on TiO 2 surfaces Tetraethoxysilane (TEOS) Rutile (1 1 0) TPD, LEED, XPS: dissociates between 200 and [656,630]

350 K to form adsorbed Si(OEt)2 and EtO; EtO decomposes at 650 K to ethylene (CH2=CH2) and H which react with EtO to form EtOH

(ethanol); SiO2 remains on the surface; pre-dosed water: increases dissociation probability, reacts

with adsorbed EtO form EtOH; Si(OEt)2 not (Et C2H5) affected bywater 178 U. Diebold / Surface Science Reports 48 (2003) 53±229

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference Vinyl-triethoxysilane Rutile (1 1 0) XPS, TPD: dissociates and produces [655]

(VTES) adsorbed Si(OEt)±(CH=CH2), and OEtads

Diethyl-diethoxy- Rutile (1 1 0) XPS, TPD: dissociates and produces adsorbed [655]

silane (DEDS) EtO; SiEt2; and EtOSiEt2 which are bound via Si±O±Ti bonds

Aminopropyl- Rutile (1 1 0) XPS, TPD: does not dissociate on TiO2(1 1 0) [655] triethoxysilane (APS)

(3,3,3-Trifluoro-propyl) Rutile (1 1 0) TPD, XPS, SSIMS: adsorbs dissociatively, forms [657]

trimethoxysilane CF3CH2CH2SiOCH3 (bound via two Si±Olattice 4 (FPTS) bonds), and ±OCH3 groups bound to Ti sites; 620 K: CF3CH2CH2± ligand decomposes through elimination of b-H (forms CF3CH=CH2 gas) or CF3 group (forms CH2=CH2 gas); ±OCH3 decomposes at 550±600 K to methane (CH4), formaldehyde (HCHO), and methanol (CH3OH); co-adsorbed water: ±OCH3 desorbs as methanol at 300 K

(g) Surveyof adsorption and photo-reactions of methylhalides on TiO 2 surfaces CH3Cl Rutile (1 1 0) Photo-oxidation with UV light: co-adsorbed [664,665] molecular oxygen and surface defects essential

for photo-oxidation; products H2CO, CO, H2O, HCl; no oxidation with hydroxyls without co-adsorbed oxygen Rutile (1 1 0) Not active for photodesorption when adsorbed [676]

byitself, photodesorbs when co-adsorbed with CH 3I CH3Br Rutile (1 1 0) Single-photon desorption process for low ¯uences, [676] substrate-mediated; laser-induced thermal desorption dominant for ¯uences 7 mJ/cm2

CD3I Rutile (1 1 0) TPD, XPS, irradiation with 254 and 334 nm photons [671] Rutile (1 1 0) Photodesorption at variable wavelength, REMPI: [672,673]

production of CD3 radicals; direct excitation of adsorbate of the primarymechanism U. Diebold / Surface Science Reports 48 (2003) 53±229 179

Table 16 (Continued ) Molecule Substrate Adsorption/reaction Reference

CH3I Rutile (1 1 0) TOF-REMPI: in multilayer ®lm C±I bonds are [670] aligned close to the surface normal in anti-parallel arrangement; at fractional monolayer coverage predominantlyin a parallel orientation with the iodine close to the surface and the Me group pointing awayfrom the surface Rutile (1 1 0) Photodesorption with 257 nm of ®lms adsorbed at [674]

90 K; photofragments: CH4,C2H6,I2,C2H2, C2H5I2,C2H5I Rutile (1 1 0) Single-photon desorption process for low ¯uences, [676] substrate-mediated; laser-induced thermal desorption dominant for ¯uences 7 mJ/cm2 Rutile (1 1 0) Post-irradiation TPD of photofragments [675] Rutile (1 1 0) Photodesorption (257 and 320 nm) in dependence [677] of thickness: Antoniewicz-type mechanism up to 1 ML; suppressed photodesorption yield between 1 and 5 ML; solvation of excited substrate electron into overlayer for coverages >5 ML and subsequent desorption from the surface Rutile (1 1 0) TPD and photodesorption; films grow stochastically [678] at 90 K, change of film morphologywith annealing

5.2.1. Carboxylic acids (formic acid, acetic acid, propanoic acid, acrylic acid, benzoic acid, bi-isonicotinic acid, oxalic acid, glycine, maleic anhydride)

5.2.1.1. Formic acid (HCOOH). Two recent reviews (Chapters 6 and 10 in [551]) give excellent summaries of the adsorption geometryand chemistryof formic acid on TiO 2, respectively. Generally, formic acid dissociates at rutile TiO2 surfaces to give formate

À HCOOHgas Olattice ! HCOOads OlatticeHads (8)

Here the H atom forms a hydroxyl with a surface atom Olattice. Quantitative information about the adsorption sites, geometry, and lateral interaction between formates and hydroxyls were obtained, see the next section. Usually, two principal overall decomposition reactions are considered to occur at elevated temperatures. Either dehydrogenation, which produces carbon dioxide and hydrogen from the original formic acid molecule

HCOOH ! CO2 H2 (9) or dehydration

HCOOH ! CO H2O (10) which renders carbon monoxide and water. The intricacies of these reaction processes at the atomic level were investigated byseveral groups [196,581±584], and are discussed below. 180 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 61. Model for formic acid adsorbed on TiO2(1 1 0). Formic acid dissociates at the surface. The resulting formate ion binds to two ®vefold coordinated Ti atoms in a bi-dentate fashion. The adsorption geometrywas mapped out with XPD and NEXAFS [586,589]. A minorityspecies adsorbed at the positions of missing bridging oxygenatoms was postulated by Hayden et al. [588].

5.2.1.2. Formate: adsorption geometry and structure

5.2.1.2.1. TiO2(1 1 0)-(1 Â 1). Adsorption at cryogenic temperature renders both, formate and (at higher coverages) molecular formic acid. The molecularlyadsorbed multilayersdesorb at 164 K [582]. Below 350 K, a saturation coverage of formate ions (0.5 ML, with 1 ML defined as one molecule per surface unit cell/per exposed fivefold coordinated Ti atom) form a regular p(2 Â 1) structure, see Fig. 61. The adsorption geometrywas mapped out in detail bycombining several photoelectron diffraction approaches [585±587]. The majorityof the HCOO À molecules are adsorbed in a bi-dentate fashion, with the oxygen atoms bridge-bonded between two fivefold coordinated Ti atoms (see Fig. 61) and the molecular plane oriented in (0 0 1) direction. A RAIRS study [588] has concluded that there is a minority species present, with the molecular plane oriented parallel to the 1 10 direction. This species is probably adsorbed at the vacancies in the bridging oxygen rows. When these minority species are included in the analysis, the results from the photoelectron diffraction measurements [585±587] are consistent with recent NEXAFS measurements [589]. The vertical Ti±O distance was determined as 2.1 AÊ and the O±C±O bond angle was estimated as 126 Æ 4. This agrees well with theoretical calculations [102,590±593]. Single formate ions can be observed with STM and non-contact AFM techniques [108,594±600]. The images are consistent with the adsorption site at the Ti atoms displayed in Fig. 61. Correlation analysis of STM images of formate ions at small coverages indicate a tendencyfor a c(4 Â 2) con®guration in addition to a p(2 Â 1) structure [599]. This maximizes the distance between CHOOÀ ions and minimizes their mutual repulsive interaction. The p(2 Â 1) structure is possiblystabilized byco- adsorbed H through reaction (8) [590,591]. Isolated formate ions are quite immobile when imaged at room temperature [108,594±600]. In an instructive STM experiment performed byOnishi and Iwasawa [598], small areas of a p(2 Â 1)-formate surface were `cleared off' (probablyby®eld desorption) by scanning with a higher bias voltage. This area ®lled up bydiffusion mainlyalong the ®vefold coordinated Ti rows, see Fig. 62. Interestingly, the boundaries moved quite uniformly, and single U. Diebold / Surface Science Reports 48 (2003) 53±229 181

Fig. 62. Serial STM images from a (2 Â 1)-formate layer on TiO2(1 1 0). Before these images were taken, a square region was rastered with a high sample bias voltage, and scans (a)±(c) were recorded 15, 26, and 35 min after the rastering. In the small area scans an isolated formate ion merged into the migrating monolayer (d)±(f). From Onishi and Iwasawa [598]. # 1996 Elsevier. formate molecules in the cleared area onlybecame mobile when hit bythe moving front of formate ions. A concerted diffusion of the H atoms adsorbed at the bridging oxygen atoms in conjunction with the formates was inferred from ®rst-principles calculations [591].

5.2.1.2.2. TiO2(1 1 0)-(1 Â 2). As discussed in Section 2.2.2, two configurations seem to co-exist for the (2 Â 1) reconstruction of TiO2(1 1 0). When 3 L formic acid were dosed on a surface with several bright strands, the (1 Â 1) terminated substrate areas were almost completelycovered bya (2 Â 1)-formate overlayer, but no formate ions were observed at the bright strands [122]. This gives credibilityto the 4 `added Ti2O3' row model that was originallyproposed in the same paper. In this model no Ti species are exposed (Fig. 18b) that would provide adsorption sites for formate. However, when formic acid was dosed on a cross-linked (1 Â 2) surface on a heavilyreduced surface

(which is thought to consist of added TiO2(1 1 0) rows interconnected by`rosette-like' structures), formate was observed to stick predominantlyto the cross-links [584]. When a (1 Â 2)-formate covered surface was ramped to higher temperature, desorption was observed as well as the formation of (1 Â 1)- terminated islands, see Fig. 63. It was proposed that oxygen atoms that are formed during the decomposition of formic acid are inserted into the lattice via reaction with interstitial Ti3 ions. The other fragments (such as CO) would desorb. In a similar high-temperature STM experiment in Iwasawa et al. [126], bright spots located on the bridging oxygen rows have been tentatively assigned as 2À carboxylate (CO3 ) reaction intermediates.

5.2.1.2.3. Modi®ed TiO2(1 1 0) surfaces. In one of the earlyadsorption studies, a stepped TiO 2(4 4 1) surface was used for adsorption of several molecules. The results were compared with a flat TiO2(1 1 0) 182 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 63. STM images of a highlybulk reduced TiO 2(1 1 0) sample covered with formic acid at room temperature during a temperature ramp of 2 K/min. The images were recorded at (A) 390 K, 52 min; (B) 420 K, 66 min; (C) 460 K, 82 min; (D) 470 K, 90 min; (E) 480 K, 97 min and ®nallyafter stabilizing the temperature at (F) 570 K, 180 min. The crystalexhibits a cross-linked (1 Â 2) reconstruction. With increasing time and temperature the number of individual bright formate features declines while small islands form within the (1 Â 2)-reconstructed terraces. The islands show a (1 Â 1) termination at 570 K. All images were taken at 0.1 nA, 1 V. From Bennett et al. [584]. # 2000 Elsevier. U. Diebold / Surface Science Reports 48 (2003) 53±229 183 crystal [545]. The (4 4 1) surface has a regular step structure which is indexed as [3(1 1 0) Â (1 1 1)] [545]. It contains a substantial fraction of Ti3 ions. The differences in work function between the stepped and the flat surfaces was attributed to inclined species at the step edges.

Reducing TiO2(1 1 0) surfaces bysputtering or electron bombardment creates more adsorption sites for formate [593]. The formate coverage scales with the reduction state of the surface, and little or no `healing' of Ti3 sites is observed upon formic acid exposure at room temperature.

Thin ®lms of TiO2 were synthesized by depositing Ti on a Ni surface and oxidizing the metal overlayer [601]. It had been shown previously [602] that an oxidized Ni94Ti6 alloygives two different surfaces depending on the preparation conditions. These exhibit either a TiO2(1 1 0) termination or a quasi-hexagonal c(2 Â 6) overlayer. HREELS from formic acid adsorbed on a four-layer TiO2(1 1 0) ®lm indicated a bi-dentate adsorption geometry( Fig. 61). In contrast, a three-layer ®lm with a quasi- hexagonal surface gives a separation between the symmetric and asymmetric OCO stretching frequencies that is closer to the one expected for monodentate bonding [601].

5.2.1.2.4. Other TiO2 surfaces. The adsorption and decomposition of formic acid on the (1 Â 3)- reconstructed surface of TiO2(1 0 0) was described byHenderson [196] and is discussed in more detail below. Formic acid dissociates as in reaction (8). No ordered LEED pattern was observed, and no details of the adsorption geometrywere given in this study. Formic acid adsorption and reaction on rutile (0 0 1) was investigated byBarteau and co-workers [583,603,604]. Again, adsorption occurs via reaction (8), with no further details on the adsorption geometrygiven.

5.2.1.2.5. Anatase. As mentioned above, studies of single-crystalline anatase surfaces are just starting to appear. The adsorption of formic acid and HCOONa was studied with DFT calculations in a slab geometrybyVittadini et al. [605].IncontrasttoTiO2(1 1 0), the most stable adsorption geometryof formate is predicted to be a molecular monodentate configuration, hydrogen-bonded to the bridging oxygen atoms. When co-adsorbed with water, it stays in the monodentate coordination, but dissociates through interaction with nearbywater molecules. Sodium formate, HCOONa, is predicted to dissociate on the dryand the water-covered surface, and to adsorb in a bridging bi-dentate geometry.

5.2.1.3. Reaction of formic acid. As described in reactions (Eqs. (9) and (10)) above, dehydration (products H2 CO2) and/or dehydrogenation (products H2O CO) are the two main reaction mechanisms of HCOOH on TiO2 surfaces. On the TiO2(1 1 0) surface, both reactions were observed [581,606] upon DCOOD exposure on TiO2(1 1 0) surfaces. Representative TPD results are displayed in Fig. 64. The desorbing D2 molecules at 400 K were attributed to a recombination of hydroxyls

OlatticeDads OlatticeDads ! 2Olattice D2;gas (11)

Moreover, a mixture of CO, CO2,D2,D2O, and DCOOD was released around 570 K. Thus both, an overall dehydration as well as a dehydrogenation reaction, would take place at this temperature. The reaction products were also measured under `catalytic conditions', i.e., at the hot surface and under formic acid pressures of 10À5 to 10À3 Pa (10À10 to 10À6 Torr). The dehydrogenation reaction (9) (resulting in D2 CO2) is dominant below 500 K. Its rate is nearlyindependent of formic acid pressure, but increases with surface temperature. The unimolecular decomposition of formate was 184 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 64. (a) TPD spectra following 3L DCOOD exposure on TiO2(1 1 0) at 230 K. Note the release of D2,D2O, CO, and CO2 around 570 K. (b) Reaction rates of formic acid on TiO2(1 1 0) simulated as a function of reaction temperature. r1: dehydration, r2: dehydrogenation; y: coverage of formates. From Onishi et al. [581]. # 1994 Academic Press. proposed as the rate-controlling step with an activation barrier of 120 Æ 10 kJ/mol. It proceeds in the following fashion

DCOOads ! COgas OformDads (12) where Oform represents an oxygen atom of formate origin. Reaction between the surface hydroxyl and a formic acid molecule from the gas phase would form D2O

DCOODgas OformDads ! DCOOads D2Ogas (13) closing the catalytic cycle. Above 500 K, dehydration occurs more rapidly than dehydrogenation. This reaction depends on the pressure, and a bimolecular reaction between an adsorbed formate and a gaseous formic acid molecule, as in

DCOOads DCOODgas ! CO2;gas D2;gas DCOOads (14) was suggested as an alternative path after decomposition of the adsorbed formate molecule. The kinetic parameters derived from these experiments were used for modeling the kinetics of the reactions, see Fig. 64b. While these results and the proposed reactions could be obtained byre-arranging the parent molecule, recent TPD studies point out that the reactions might be more intricate, and the substrate U. Diebold / Surface Science Reports 48 (2003) 53±229 185

14 2 Fig. 65. TPD spectra from a formic acid exposure of 4:8 Â 10 molecules/cm (monolayer saturation) to the TiO2(1 0 0)- (1 Â 3) surface at 170 K. Note that most of the H2O desorption is completed before the onset of CO desorption. From Henderson [196]. # 1995 The American Chemical Society. itself might playa major role in the decomposition of formate. A TPD/SSIMS study [196,582], combined with 18O labeling of either the surface (byreplacing most of the surface 16O atoms by 18O), or the adsorbate (using 18O-containing reactants), is quite instructive in this respect. 4 The results on the microfaceted TiO2(1 0 0)-(1 Â 3) surface are considered ®rst [196]. Theyshow the same general trend as similar experiments on the (1 1 0) surface [582]. Results from TPD desorption products after exposure of a monolayer of formic acid on TiO2 at 170 K are displayed in Fig. 65. The major reaction product is CO, which desorbs around 550 K in a near-®rst-order process. While this points towards a dehydration reaction, it is puzzling that desorption of water is nearly completed before CO or other reaction products leave the surface. It should be pointed out that adsorbed CO is not stable on TiO2 at these high temperatures (see Section 5.1.4), i.e., the amount of desorbing CO is reaction-limited. This means that the majorityof the desorbing water is formed in a process that is independent from the decomposition of the formate ions. Verysimilar results (i.e., CO

4 The {1 1 0}-faceted model for the TiO2(1 0 0)-(1 Â 3) surface was adopted in [196], see Fig. 23b. As discussed above, recent measurements seem to indicate that this simple model might not be valid. However, this does not affect the considerations about formic acid decomposition in a major way. It is sufficient to assumeÐand well supported by experimental evidenceÐthat this surface contains reduced Ti3 ions and that twofold coordinated, bridging oxygen atoms are present as well. 186 U. Diebold / Surface Science Reports 48 (2003) 53±229 desorption as the major reaction product at 550 K, where most of the H2O has alreadyleft the surface) was observed on a TiO2(1 1 0) surface [582]. One possibilityis that water is formed when two of the surface hydroxyls that have formed during the dissociative adsorption of formic acid (Eq. (8)) combine as

2OlatticeH ! H2Olattice g Olattice VO (15) For example, the two OH groups that are sketched in Fig. 61 on a rutile (1 1 0) surface could recombine and leave the surface as water. This would cause the creation of a vacancyV O, i.e., a (temporary) reduction of the surface. HREELS measurements on TiO2(1 1 0) point in this direction [582]. 18 As expected from this scenario, a substantial fraction of water desorbs as H2 O desorbing from an 18O-enriched surfaces.

As seen from Fig. 65, formaldehyde, H2CO, is also a major desorbing species from TiO2(1 0 0)- (1 Â 3). Barteau and co-workers [583,603] pointed out that reduced Ti species are capable of promoting the formation of formaldehyde from formic acid. It can be produced either in an unimolecular reaction

HCOOads Hads VO ! H2COgas Olattice (16) or in a bimolecular reaction

2HCOOads VO ! H2COgas CO2;gas Olattice (17)

The reaction in Eq. (17) is the dominant decomposition channel on the `{1 1 4}-faceted' TiO2(0 0 1) surface which contains substantial amounts of fourfold coordinated Ti ions [603]. Because of the 3 presence of undercoordinated Ti ions on TiO2(0 0 1)-(1 Â 3), it is not too surprising that H2CO is a major product of formate decomposition at this surface as well. However, it was found byHenderson

[582] that a small amount of formaldehyde was also produced on a TiO2(1 1 0) surface, in contrast to the results byOnishi et al. displayedin Fig. 64. Possibly, the temporary oxygen vacancies, created by reaction (15) playa role in this process. One oxygen atom results from either reaction (16) or (17). It could be incorporated into the lattice at vacancysites produced bywater desorption ( Eq. (15)). An alternative scenario is suggested bythe STM images in Fig. 63. The images show the substantial oxidation capacityof formic acid at elevated temperatures. This means that the resultant O atom could also react with Ti interstitials from the reduced bulk. The next product displayed in the TPD spectra in Fig. 65 is formic acid. (The HCO fragment in

Fig. 65 is a cracking product of H2CO and HCOOH in the mass spectrometer.) Multilayers evolve when higher formic acid dosages are used. The trace in Fig. 65 shows two features, and the one at higher temperatures coincides with the release of CO from the surface. Possibly, decomposition of formate results in the release of protons that can lead to a recombination with adsorbed formate

HCOOads ! OHads COgas (18)

HCOOads OHads ! HCOOHgas Oads (19) The reactions described in Eqs. (15)±(19) leave open an interesting question, however. When decomposition of formate results mainlyin evolution of CO and is decoupled from the formation of water, what happens with the hydrogen? Especially on the TiO2(1 1 0) surface, the relativelysmall U. Diebold / Surface Science Reports 48 (2003) 53±229 187 amounts of hydrogen-containing species (HCOOH, H2O, H2CO, and C2H2) do not balance the amount of CO produced [582]. It was speculated that the excess H diffuses into the bulk [196]. Such a process was observed on ZnO 0001 [607], but this possibilityhas not yetbeen explored for TiO 2. Little or no CO2 production was observed in the TPD spectra in Fig. 65, again in contrast to Onishi's results. (The trace in Fig. 65 comprises also the mass spectrometer cracking fragments from desorbing formic acid and other molecules.) In addition, no H2 evolution was observed on either the TiO2(1 0 0)- (1 Â 3), the TiO2(1 0 0)-(1 Â 1), or the TiO2(1 1 0)-(1 Â 1) surface. Thus the overall dehydrogenation reaction (Eq. (9)) was not observed in Henderson's experiments [196,582].

5.2.1.4. Formic acidÐconclusion. The adsorption of formic acid is one of the best-investigated organic systems at this point. The adsorption geometry of the formate molecule on TiO2(1 1 0) is sufficientlywell known for more intricate studies, such as trying to understand the complexities of the diffusion process [590,591]. The fact that it can be observed directlywith scanning probe techniques has also contributed to the understanding of this adsorption reaction. However, it is unclear at this point what causes the different results for formic acid decomposition/reaction that were obtained bydifferent groups using virtuallythe same systems and techniques. It is now clear that the substrate plays a very active role in the decomposition and reaction processes; surface atoms are incorporated in the reaction products, and the substrate itself is re-oxidized upon formate decomposition. It would be well worth an investigation what high-temperature processes occur under which conditions, and in how far the reduction state of the

TiO2 sample itself plays a role.

5.2.1.5. Acetic acid (CH3COOH). The adsorption of acetic acid generallyfollows the same trends found for formic acid. Far fewer investigations were performed on this molecule, however. À At room temperature on TiO2(1 1 0), acetic acid adsorbs as acetate H3CCOO in a (2 Â 1) structure, with the two O molecules bonded in a bridging (bi-dentate) con®guration across two ®vefold coordinated Ti atoms with the C±C bond perpendicular to the surface. This puts the H atoms in the methyl group parallel to the surface plane. ESDIAD measurements show two contributions in the H emission. A lobe peaked at normal emission is surrounded bya ring. The H ions in the normal emission are attributed to surface hydroxyls that are formed during the dissociative adsorption, similar to reaction (8) for formic acid. The H ions in the ring most likelystem from the ±CH 3 moietyof the acetate molecule that freely rotates around the molecular axis. NEXAFS experiments also indicate that the molecule stands upright, with an overall twist angle of the molecular plane of 26 Æ 5 from the [0 0 1] direction [589]. Adsorption of acetic acid on the (1 Â 2) surface does not modifythe LEED pattern, but ESDIAD reveals H desorption with a weaker off-normal contribution consistent with the Ti2O3 model of the reconstruction. The overall coverage on a (1 Â 2)-reconstructed surface is also smaller [608]. The decomposition of acetic acid was directlyobserved with STM byOnishi et al. [609]. STM images after a temperature jump to 580 K were analyzed. The number of bright spots on the surface decreased exponentiallywith time. This was assigned as a unimolecular decomposition of acetate to release ketene

À À CH3COOads ! H2CCOgas OHads (20) Plotting of the number of the bright spots in consecutive STM images against time on a semi-log scale results in a rate constant of 4 Æ 1Â10À3 sÀ1. This agrees well with a rate law deduced from a thermal desorption study [604]. In addition to the acetate molecules, a few immobile, bigger species 188 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 66. Selectivityof CH 3COOH decomposition on differentlypretreated TiO 2(0 0 1) surfaces. The selectivityis de®ned as the fraction of total carbon appearing in each product. From Kim and Barteau [604]. # 1990 Academic Press. were observed and assigned as carbonaceous residues formed in a side reaction. These disappeared at higher temperatures. It is possible to distinguish between co-adsorbed formate and acetate molecules with non-contact AFM. The latter appear higher in the images [600,610]. Acetate molecules at the boundarybetween different (2 Â 1) domains were observed to be mobile during scanning with the AFM tip [611]. The ¯uctuations were explained invoking interactions between hydroxyl H atoms and acetate.

A comprehensive TPD/XPS studyon differentlytreated TiO 2(001)surfaceswasperformedbyKim and Barteau [604]. The surface structure and the degree of reduction after different pretreatments is depicted in Fig. 66. At 390 K, approximately15% of acetates desorb via recombinative adsorption as acetic acid, together with some water. Decomposition occurred at higher temperatures via three different pathways as summarized in Fig. 66. On the sputtered, oxygen de®cient surface (``no LEED'' in Fig. 66), the main decomposition product is atomic C that is then burned off as CO via reaction with surface oxygen. On a more oxidized surface, ketene (CH2CO) is produced via a net dehydration reaction

CH3COOH ! H2CCO H2O (21) that peaks in selectivityat the `{0 1 1}-faceted' surface. On the `{1 1 4}-faceted' surface, production of another decomposition product, acetone (CH3COCH3) was observed. This surface might contain Ti cations in several undercoordinated sites and production of acetone maytake place through a bimolecular reaction where two acetate molecules are coordinated to a common Ti4 surface cation. U. Diebold / Surface Science Reports 48 (2003) 53±229 189

5.2.1.6. Propanoic acid (C2H5COOH). Both the adsorption on rutile (1 1 0) and the reaction on rutile (0 0 1) surfaces occur analogous to other acetic acid reactions. Adsorption of propanoic acid probably forms propanoates which order in a (2 Â 1) overlayer. The bond geometry in NEXAFS experiments indicate an upright molecule [589]. Two main decomposition products are formed at elevated temperatures. Methyl ketene (CH3HC=CO) forms through unimolecular dehydration on `{0 1 1}- faceted' TiO2(0 0 1) surfaces. TPD data on the `{1 1 4}-faceted' surface were consistent with formation of 3-pentanone (di-ethyl ketone, CH3CH2COCH2CH3) [604].

5.2.1.7. Acrylic acid (CH2=CHCOOH). TPD, XPS, and scanning kinetic spectroscopywere employed to studythe kinetics of acrylicacid decomposition on TiO 2(0 0 1) [612]. In the latter technique, the crystal is pre-dosed as in a regular TPD measurement, but a flux of molecules (provided by a background pressure of the order of 5 Â 10À9 Torr in the reported experiment) is kept above the sample surface. This allows the observation of high-temperature reaction paths that are often not accessible in a regular TPD À experiment. Most likely, adsorption occurs dissociatively as acrylate (CH2=CHCOO ) and is little affected bythe CH 2=CH± (olefin) side chain. A range of reaction products due to the olefin fragment can be seen that are absent in the case of propanoic acid decomposition [604]. These include C4 products (butene and butadiene) and benzene. The formation of di-vinyl ketone

is onlyseen on the `{1 1 4}-faceted' surface [604].

5.2.1.8. Benzoic acid (C6H5COOH). An ESDIAD/LEED studyof benzoic acid adsorption at room temperature indicates that it adsorbs dissociativelywith the two oxygenatoms bridge-bonded to fivefold coordinated Ti atoms [613]. The ESDIAD H pattern was interpreted as an upright benzoate molecule with an upright phenyl ring such as in

STM images showed a dimerization of the benzoate rings along the 1 10 direction [614].

5.2.1.9. Bi-isonicotinic acid. The adsorption of more complex acids was extended to bi-isonicotinic acid (2,20-bipyridine-4,40-dicarboxylic acid) in an XPS/NEXAFS study, combined with quantum chemical

(INDO) calculations [615]. This molecule is the ligand that anchors organometallic dyes to TiO2 nanoparticles in electrochemical applications such as in the `GraÈtzel cell' [29], see Section 5.3.2. The proposed adsorption geometry, derived from the calculations and consistent with the experimental data, is shown in Fig. 67. The bonding follows in principle the trend expected from formate and benzoate adsorption. The molecule binds to two neighboring Ti rows in a bi-dentate fashion through the dehydroxylated O atoms, and is connected via the rings along the 110 direction. N1s NEXAFS spectra were modeled successfullyusing this adsorption geometryin a cluster model [616]. 190 U. Diebold / Surface Science Reports 48 (2003) 53±229

Fig. 67. Calculated structure of bi-isonicotinic acid adsorbed on TiO2(1 1 0). From Persson et al. [616]. # 2000 The American Chemical Society.

5.2.1.10. Oxalic acid (HOOC±COOH). Fahmi et al. [617] performed theoretical investigation of oxalic acid (a planar molecule with the two formate groups linked together in two possible configurations). The study was motivated by the observation that crystal growth of the two polymorphs of titanium dioxide, rutile and anatase, are stronglyaffected bythe presence of oxalic acid. The growth of rutile is enhanced, suggesting that it binds to anatase more strongly. The surfaces were modeled with `polymers' (essentially Ti atoms surrounded bya square of oxygenatoms, and arranged either in a linear `rutile' or a square `anatase') fashion. The calculations suggest that oxalic acid undergoes dissociation, i.e., a splitting of 2À both H atoms. The resulting oxalate ion, C2O4 , is bonded through the oxygen atom to neighboring Ti atoms.

5.2.1.11. Glycine (NH2CH2COOH). Multilayers of glycine, adsorbed on TiO2(1 1 0), were studied with synchrotron radiation-based UV light [618,619]. This linear molecule consists of three groups and two configurations are possible, as in

(22) U. Diebold / Surface Science Reports 48 (2003) 53±229 191

UPS spectra of shallow core and valence levels indicate that it adsorbs in multilayers of the zwitterionic ion where the H atom from the carboxyl group transfers to the amino group to form an NH3 radical. The resulting molecule is bi-polar. Photon damage of the glycine layers occurs fast (cross-section of 5:4 Â 10À16 cm2, comparable with electron excitation in the gas phase). Small coverages on the

TiO2(1 1 0)-(1 Â 2) surface indicate dissociative adsorption where most of the amino group is released into the gas phase.

5.2.1.12. Maleic anhydride. Maleic anhydride is not a carboxylic acid; but it is still listed here as it exhibits some similarities in its adsorption and reaction pathways at the TiO2(0 0 1) surface [620]. Adsorption at a TiO2(0 0 1) surface was modeled with semi-empirical calculations of a relaxed Ti13O45H38 cluster. Of the considered adsorption geometries, the most stable configuration was dissociative adsorption. A model was suggested where one C±O±C bond was broken and bonded to one Ti and a neighboring oxygen atom in the following fashion

(23)

The observed reaction products (see Table 16) were rationalized with the following scheme:

(24)

Here the adsorbed molecule maydissociate via two pathways,either parallel or perpendicular to the molecular c-axis. The TPD data suggest breaking of the C=C bond and formation of ketene with a hydrogen atom that could either come from more complete dissociation of maleic anhydride or as an impurityfrom the bulk. A cut of the molecule in perpendicular direction would result in the formation of acetylene as well as CO and CO2. Formation of C4 and C6 products could evolve from coupling of acetylene adsorbed on surface defects.

5.2.2. Alcohols (methanol, higher alcohols)

5.2.2.1. Methanol. The adsorption of methanol was studied on various TiO2 (rutile) surfaces (Table 16(b)). Experiments and theoryagree that both, molecular and dissociative adsorption take place. Dissociation happens via breaking of the O±H bond of CH3OH, leading to an adsorbed methoxy species and, probably, a hydroxylated bridging oxygen atom. Several adsorption configurations are separated bya verysmall or maybeeven negligible barrier at room temperature. Interestingly,only 192 U. Diebold / Surface Science Reports 48 (2003) 53±229 molecular desorption was observed on TiO2(1 1 0), even when point defects are present. In contrast, a varietyof reaction products form on TiO 2(1 0 0).

5.2.2.1.1. Methanol on TiO2(1 1 0). The most extensive and recent adsorption studyof methanol on the rutile (1 1 0) surface was performed byHenderson et al. [151,621]. An earlier studybyOnishi et al. [545] reported molecular adsorption at room temperature; an ordered overlayer was not observed. Henderson's results also showed that the majorityof molecules adsorbs intact, and leaves the surface around 295 K. Methanol is the onlycarbon-containing species in TPD spectra, and no evidence for residual decomposition products was observed in repeated TPD experiments. The TiO2(1 1 0) surface in this experiment contained about 8% vacancies [621], and a TPD peak at 480 K was attributed to recombinative adsorption of dissociated products at vacancysites. Isotopic scrambling TPD and SSIMS experiments showed that methanol desorbing at 350 K is possiblyformed via recombinative desorption from dissociated methoxyspecies at non-vacancysites. Monolayercoverage is reached around 3:4 Â 1014 molecules/cm2. This is consistent with XPS results after room-temperature adsorption [622]. A weak, streakyLEED pattern ( Fig. 68) is consistent with a coverage of 2/3 ML, and the model displayed in Fig. 68 can rationalize the observed results for monolayer coverages [621]. Higher exposures lead to CH3OH molecules hydrogen-bonded to the bridging oxygen atoms and multilayers. First-principles calculations were performed using densityfunctional theoryand pseudopotentials [623]. The rather powerful VASP code [624±626] was applied in both the static and dynamic mode. Three adsorbed species were considered at different coverages and con®gurations; molecularly adsorbed CH3OH, surface methoxy(formed byO±H bond scission), and species formed via C±O bond À scission, i.e., a CH3 ion and an OH group. (In the gas phase it is far easier to break the latter bond, however, the calculations show that this process appears to be activated at the TiO2(1 1 0) surface.) The adsorption energyof molecular CH 3OH and the different dissociative species is similar. Repulsive and attractive intermolecular interactions playa major role, which leads to a coverage dependence for the stable con®guration. Spontaneous O±H bond breaking was observed in molecular dynamics calculations of adsorbed CH3OH. In this sense, the model in Fig. 68 must be regarded as one of several possibilities. The tendencyfor molecular vs. dissociative adsorption, and the role of H-bonding, was compared to other R±OH molecules in [102]. Co-adsorbed water showed weak to no changes in the surface chemistryof methanol [621]. Co- adsorbed oxygen opens up new decomposition channels. Oxygen exposure to a slightly defective

TiO2(1 1 0) surfaces leads to dissociation of the O2 molecules which results in O adatoms. This leads to additional O±H bond cleavage, and the formation of H2CO above 600 K. When O2 molecules were adsorbed at 150 K, an additional low-temperature desorption peak of H2CO was observed, possibly formed via an abstraction of an H atom from an adsorbed methoxyspecies. A high cross-section for electron-stimulated desorption/decomposition was reported [151].Thisis probablythe reason whythe LEED pattern in Fig. 68 was very``fragile'' and faded after a few seconds [621], and whyno LEED pattern was observed in other studies. In ESD experiments [151], surfaces were prepared to contain the saturated monolayer, methoxy species only, and methoxy species adsorbed at vacancies only. Electron bombardment of these different systems consistently showed high cross-sections for decomposition, and a varietyof products, depending on coverage, adsorption state, and adsorption site.

5.2.2.1.2. Methanol on TiO2(0 0 1) and TiO2(1 0 0). The adsorption and thermal desorption of methanol on differentlypretreated TiO 2(0 0 1) surfaces was studied byKim and Barteau [627]. At 200 K, methanol U. Diebold / Surface Science Reports 48 (2003) 53±229 193

Fig. 68. LEED patterns and proposed overlayer structure of methanol adsorbed on TiO2(1 1 0). (A) Clean TiO2(1 1 0) surface with 8% vacancies in the bridging oxygen rows, (B) LEED patterns of a monolayer of CH3OH, (C) same but in off-normal geometry. (D) Model of CH3OH adsorption based on LEED, TPD, and SSIMS results. Dissociative adsorption at oxygen vacanciesassurfacemethoxies(DV) blocks nearbyadsorption sites. A poorlyordered, mixed layerof 1/3 dissociated methoxy species (D), bridge-bonded to two ®vefold coordinated Ti atoms, and 2/3 molecular methanol (M), in on-top geometry, could account for the 3 Â n LEED pattern in B and C. From Henderson et al. [621]. # 1999 Royal Society of Chemistry. 194 U. Diebold / Surface Science Reports 48 (2003) 53±229 adsorbs both dissociativelyand molecularly,with molecular CH 3OH desorbing below room temperature. This is consistent with UPS results [628]. Half of the dissociated methoxyspecies were removed via recombinative desorption as methanol. The other half left the surface in a varietyof reaction products (see Table 16(b)) at higher temperatures. The selectivityfor the product formation depends on the degree of undersaturation of the substrate surface atoms. Methane is thought to be produced bydeoxygenationof the adsorbed methoxygroup on oxygenvacancies and was found to occur with the highest selectivityat the `{0 1 1}-faceted' surface. Dimethyl ether (CH3±O±CH3) was observed onlyat the `{1 1 4}-faceted' surface, and probablyforms via disproportionation of pairs of methoxides coordinated to fourfold coordinated Ti cations. On the sputtered surface, methoxides adsorbed at 300 K decompose into CO. The influence of point defects on methanol adsorption was investigated byRomaÂn et al. [628]. On both, the

TiO2(1 1 0) and the TiO2(1 0 0) surface, the methanol coverage increased with the number of defects created byelectron or Ar bombardment [628].

5.2.2.2. Higher alcohols. The adsorption and photo-reaction of 2-propanol on TiO2(1 1 0) and TiO2(1 0 0) was studied byEngel and co-workers and is discussed in Section 5.3.3. The adsorption and reaction of ethanol as well as higher aliphatic alcohols (n-propanol, iso-propanol) on TiO2(0 0 1) was studied byKim and Barteau [629]. Ethanol adsorbs at 200 K in a mixed layer as molecular C2H5OH and dissociated ethoxyspecies C 2H5O. The molecular ethanol desorbs below 300 K, and adsorption at room temperature leads to an ethoxylated surface. About half of the ethoxy species recombine with a proton

CH3CH2Oads Hads ! CH3CH2OHgas (25) and desorb as ethanol. The other half decomposes to produce either acetaldehyde

CH3CH2Oads ! CH3CHOgas Hads (26) or ethylene

CH3CH2Oads ! H2CCH2 Olattice Hads (27) The adsorption of the two isomers of propanol on the same surface is qualitativelysimilar to the one observed for ethanol [629]. Co-adsorbed hydroxyls are obviously important for the recombination reaction (Eq. (25)). Water adsorption on a TiO2(1 1 0) surface covered with ethoxyspecies was investigated byGamble et al. [630]. The surface ethoxys were prepared by dissociative adsorption of either deuterated ethanol or tetraethoxysilane (TEOS). Two different species were identi®ed. One that recombined with surface hydroxyls in the temperature range 250±400 K, and desorbed as ethanol gas (Eq. (25)). This species were attributed to an ethoxygroup adsorbed on a ®vefold coordinated Ti atom. The second species was attributed to bind to oxygen vacancies, similar to the methoxy species in Fig. 68. This one binds to two undersaturated Ti atoms and decomposes at 650 K, giving ethylene (Eq. (27)) and ethanol. These results are in good agreement with studies of hydroxylated rutile powder materials [631].

5.2.3. Aldehydes (RCHO) and ketones (RCOCH3) (formaldehyde, acetaldehyde, benzaldehyde, acetone, acetophenone, p-benzoquinone, cyclohexanone, cyclohexenone) Three different aldehydes with increasing complexity were investigated by Barteau and co-workers

[632±636] (Table 16(c)). All the experiments were performed on TiO2(0 0 1) single crystals, either right U. Diebold / Surface Science Reports 48 (2003) 53±229 195 after sputtering (``reduced surfaces'') or after annealing (oxidized surfaces). Reductive coupling to dimers of the parent molecule was observed on reduced surfaces for the higher aldehydes.

5.2.3.1. Formaldehyde. Adsorption of formaldehyde, HCHO, was studied in [490,632]. Upon reaction on stoichiometric TiO2(0 0 1) surfaces, formation of methanol is observed which leaves the surface together with formaldehyde. It is thought that the formation of methanol occurs in a Cannizzaro-type [637] reaction. In this reaction, aldehydes disproportionate when treated with concentrated NaOH or other strong bases. One molecule of aldehyde oxidizes another one to the acid and is itself reduced to the primaryalcohol,

À OH À 2RCHO ! RCH2OH RCOO (28)

In the case of TiO2, it was suggested that the formaldehyde adsorbs on a cation through s lone pair donation from the oxygen of the carbonyl

in agreement with the observations in [490]. Under incorporation of an oxygen atom from the

TiO2 substrate it would then form a dioxymethylene species. Hydride transfer from one dioxymethylene to a neighboring one would give one methoxide, which later desorbs as a methanol, and a formate ion.

On reduced TiO2 surfaces, additional desorption of methanol and formaldehyde was observed around 550 K. The formaldehyde probably decomposes completely to adsorbed carbon, hydrogen, and oxygen

HCHOads VO ! 2Hads Cads Olattice (29) which is incorporated into the lattice and partiallyoxidizes the surface. The H is consumed byreduction of another formaldehyde molecule to a methoxide species

HCHOads Hads ! CH2OHads (30) which then desorbs as methanol. The adsorbed carbon, Cads, that is formed through reaction (29) is oxidized to CO and CO2. Thus, the formation of methanol on the reduced surface does not involve formation of formate as in the Cannizzaro reaction.

5.2.3.2. Acetaldehyde. In contrast to formaldehyde, acetaldehyde, H3CCHO was found to couple to C4 products on TiO2(0 0 1) [633,634]. The C±C bond formation is relativelystructure insensitive, as it occurs on `{1 1 4}'- and `{0 1 1}-faceted' surfaces. Primaryproducts are either crotonaldehyde

2CH3CHO ! CH3CHCHCHO H2O (31) or crotyl alcohol

2CH3CHO ! CH3CHCHCH2OH Oads (32) 196 U. Diebold / Surface Science Reports 48 (2003) 53±229

Cannizzaro disproportionation into acetate plus ethoxides is a minor reaction channel. On sputter- reduced surfaces, reductive coupling to form butene

2CH3CHO ! CH3CHCHCH3 2Olattice (33) was the dominant reaction pathway.

Thermallycreated point defects on TiO 2(1 1 0) cause an extraction of the O atom from the adsorbed formaldehyde molecule to yield C2H4 [490]. No deoxygenation reaction was observed on stoichiometric surfaces.

5.2.3.3. Benzaldehyde. The adsorption and thermal reaction of benzaldehyde on Ar sputtered, reduced TiO2(0 0 1) surfaces produces stilbene, two phenol rings coupled via a C=C double bond [635]. This reductive coupling occurs onlyin the presence of undercoordinated Ti atoms. The stilbene yield decreased by an order of magnitude when the TiO2(001)surfacewasoxidizedbyannealing to 850 K prior to benzaldehyde adsorption. NEXAFS measurements were performed to determine the reaction intermediates [636]. A benzaldehyde-covered surface was progressively heated to higher temperatures, and Ti L-edge, as well as O and C K-edge spectra were taken. It was concluded that all of the oxygen originally contained in the carbonyl groups is donated to the surface by 300 K, and that the phenyl rings are oriented 548 with respect to the surface. These rings become more parallel as stilbene is produced.

5.2.3.4. Acetone and acetophenone. Several ketones (RCOCH3) were investigated byPierce and Barteau [638] on TiO2(0 0 1) surfaces. Acetone (where R H3) and acetophenone (where R phenyl ring) molecules were adsorbed either directly after sputtering, or after annealing of the sputter-reduced substrate. On the reduced surface, the primaryreaction is the dimerization to a symmetric olefin with twice the carbon number of the reactant, see Table 16(c). For acetone, the main product is 2,3- dimethyl-2-butene, for acetophenone 2,3-diphenyl-2-butene. Very high conversion (up to 90% for acetophenone) and selectivitywas found for carbonylcoupling. This reaction and the rate of activity and selectivityis comparable to the formation of stilbene from benzaldehydementioned above. In all cases, the yield for C±C coupling was greatly reduced on more stoichiometric surfaces.

5.2.3.5. Cyclic ketones. In addition to acetophenone, several other cyclic ketones were investigated byIdriss and Barteau [639,640] on reduced TiO2(0 0 1) surfaces, see Table 16(c). For p-benzoquinone coupling reactions were observed to bi- and terphenyl. XPS measurements indicated that even higher oligomers are formed. These are non-volatile and remain on the surface at elevated temperatures. Like the monofunctional alkynes and ketones, reductive coupling of p-benzoquinone requires Ti cations in oxidation states lower than 4. Other reaction products include unimolecular reduction to benzene and phenol.

Cyclohexanone and cyclohexenone also couple reductivelyto C 6H10=C6H10 and

C6H8=C6H8 compounds, respectively. In each case, the reduced surface becomes oxidized by incorporation of the resulting O atoms into the lattice.

5.2.4. Cyclo-trimerization of alkynes (RCBCH) on reduced TiO2 surfaces and related reactions A veryinteresting class of reactions are the trimerization of alkyneson reduced TiO 2 surfaces to form aromatic compounds (Table 16(d)). This process was investigated thoroughlybyBarteau and U. Diebold / Surface Science Reports 48 (2003) 53±229 197

Scheme 1. co-workers [641±647] and was found for a number of alkynes. In all cases, this reaction occurred only on reduced surfaces (produced bysputtering) that contained substantial amounts of Ti 2. The reaction scheme for the most simple molecule, acetylene, HCBCH, is thought to occur as in Scheme 1. In Scheme 1, two acetylene molecules would bind to such an undercoordinated Ti2 site and form a

®ve-member ring (a metallocyclopentadiene) consisting of C4H4 and the lattice Ti ion, which becomes oxidized in the cyclization process. A third acetylene molecule is incorporated into the ring, forming benzene, which desorbs at around 400 K [641]. During benzene formation, the substrate atom is reduced to Ti2, and the catalytic cycle is closed. Other reactions include dimerization to dienes (e.g. propene, H2C=CH±CH=CH2, in the case of acetylene in Scheme 1), and unimolecular hydrogenation to ole®ns (e.g. acetylene to ethene, CH2=CH2). Fig. 69 shows the in¯uence of the reduction state of the substrate on the activityof the TiO 2 sample, in this case for methylacetylene, CH3CBCH. After sputtering, the yield for trimerization to trimethyl benzene (an aromatic ring with three methyl groups, see Table 16(d)), dimerization, and hydrogenation of the parent molecule is highest. XPS showed that annealing above 600 K oxidizes the TiO2 surface; and that the yields for all reaction products decrease dramatically when the surface becomes oxidized [641].

Fig. 69. Molar yield of products from TPD of methylacetylene, CH3CBCH, as a function of annealing temperature after Ar ion bombardment of a TiO2(0 0 1) surface. The yields were scaled so that the total yield of a carbon-containing species from the Ar-bombarded surface (300 K) sums to unity. For annealing temperatures above 600 K, the substrate re-oxidizes, and the activityfor cyclotrimerizationdecreases. From Pierce and Barteau [641]. # 1994 The American Chemical Society. 198 U. Diebold / Surface Science Reports 48 (2003) 53±229

An analysis of cracking patterns of different trimethyl benzenes in the quadrupole mass spectrometer indicated that the 1,2,4 isomer is the most abundant species (see Table 16(d)). This implies a random insertion of acetylenic units (B) into the aromatic ring. Similar reactions were found for higher alkynes, basically acetylene molecules functionalized with groups with increasing complexity(see Table 16(d) for the reactants and cyclic products). It even occurs for a hetero-atom containing alkyne, trimethylsilylacetylene, albeit not with a very high selectivity [644]. The validityof reaction in Scheme 1 was tested byadsorption/thermal desorption of two molecules, allene, and cyclooctatetraene. Thermal desorption of allene (CH2=C=CH2), an isomer to methylacetylene, CH3CBCH, [645] showed as the principal product propylene, CH3CH=CH2, through hydrogenation. For sputtered TiO2 surfaces three dimeric products were observed (dimethylene cyclobutane, benzene, and an open-chain C6H10 dimer, see Table 16(d)). No formation of trimethylbenzene, or any other trimer product was observed, indicating that an unsaturated triple bond is necessaryfor the trimerization reaction. An alternative intermediate to the metallocyclopentadiene (Scheme 1) has been tested in [647]. It is conceivable that a pair of C4 ligands combine to form a C8 ligand, cyclooctatetraene (COT ) whichwouldthendecomposetobenzeneandaC2fragment.WhenCOTwasadsorbedonareduced

TiO2 surface, it did indeed produce benzene, albeit at a higher temperature than the one where formation of aromatic rings bytrimerization of acetyleneis observed. In addition, the formation of cyclooctatriene was observed to occur at the same temperature as benzene formation, pointing to the same metallacycle reaction intermediate. This makes COT an unlikely intermediate for the alkyne trimerization. Recent NEXAFS measurements analyzed the probable adsorption geometry of COT on the sputter-reduced surface [646].

5.2.5. STM of pyridine, its derivates, and other aromatic molecules (pyridine, 4-methylpyridine, benzene, m-xylene, phenol) As pointed out several times throughout this review, STM is uniquelysuited to directlymonitor and visualize surface adsorption reactions. In addition to the studies on formate and higher organic acids, mentioned above, adsorption of pyridine and its derivates were studied with STM (Table 16(e)). Iwasawa's group is the foremost leader in this respect, and most of the examples reviewed here are drawn from this group. It should also be mentioned that one of the ®rst STM images of an aromatic molecule was obtained byimaging phenol on a rutile (1 1 0) surface in air [648,649]. Pyridine is a Lewis base (the N atom in the aromatic ring acts as an electron donor), and is typically used for titration of acidic sites in catalysts. The molecule is thought to adsorb on electron acceptors, i.e., the exposed cations on metal oxides. Hence, the molecule is expected to bind with the N-end down to the Ti atoms on the TiO2 surface. However, a combined XPS/TPD/STM studyshowed that pyridine is generallyonlyphysisorbedon TiO 2(1 1 0) [650]. In this studyit was found that the N1s line of pyridine does not show a chemical shift in XPS. In addition, the TPD spectra of pyridine and benzene (the aromatic ring without the N atom) are rather similar, indicating the absence of a N±Ti bond. In order to test this, TPD was performed on adsorbed 2,6-dimethylpyridine (2,6-DMP, with two CH3 groups attached to the C atoms next to the N, see Table 16(e)) and m-xylene (the same con®guration, but with the absence of the N in the ring). The rather similar TPD results for all these molecules indicate that electrostatic van-der Waals interactions playa major role in the adsorption of all U. Diebold / Surface Science Reports 48 (2003) 53±229 199 molecules. Molecular dynamics simulations support this view, and indicate that a ¯at adsorption geometryfor both, pyridineand benzene adsorption. This is in agreement with an electron spectroscopy studyof benzene on TiO 2(1 0 0)-(1 Â 1) [651]. In agreement with the weak bonding of pyridine, atomically resolved STM images from a TiO2(1 1 0) surface show that the molecule is rather mobile [124,650]. However, pyridine binds more strongly to steps with certain orientations. An analysis of step directions shows that only steps where fourfold coordinated Ti atoms reside are active for strong pyridine adsorption at room temperature [124]. Even there, an exchange between pyridine molecules adsorbed on terraces and at step edges is frequently observed. The 2,6-DMP was somewhat less mobile in STM [650]. Pyridine also seemed to desorb during scanning at room temperature [650]. A thermallyactivated adsorption of a minorityspecies was observed in STM measurements under a pyridine background pressure of 1 Â 10À6 Pa (1 Â 10À8 mbar) and a sample temperature of 350 K [652]. Under these conditions, particles were observed to form and grow at the surface. It was proposed that stable species are formed, possiblyvia dehydrogenation, which then would act as nucleation sites for the growth of larger pyridine condensates.

DFT of pyridine on TiO2(1 1 0) [653] showed that the most stable adsorption state is the upright con®guration with the 2,6-hydrogen atoms interacting with bridging oxygen atoms via a hydrogen- bond-like interaction. A ¯at-lying pyridine was found with a low surface diffusion barrier along the [0 0 1] direction. Conceivably, these correspond to movable species observed by STM. In agreement with STM results [124,650], fourfold coordinated Ti atoms obtained byremoval of two adjacent bridging oxygen atoms were found to be stronger Lewis acid sites than ®vefold coordinated Ti atoms.

STM experiments with 4-methylpyridine (4-MP, with a CH3 group attached opposite to the N-end of the aromatic ring, see Table 16(e)) seem to con®rm the idea of a mobile, ¯at lying, and less mobile, upright species [654]. When this molecule adsorbs in an upright position, it should stick out from the surface much higher than a ¯at-lying species, giving rise to a higher topographic contrast than pyridine. Indeed, three species were identi®ed in STM, and were assigned as upright, ¯at immobile, and ¯at mobile 4-MP molecules The ¯at, mobile species was the majorityspecies. Both, the upright and the ¯at immobile 4-MP's were attributed to being adsorbed at sites next to a bridging oxygen vacancy.

5.2.6. Adsorption and reaction of silanes (RSiX3) (TEOS, diethyldiethoxysilane, vinyltriethoxysilane, aminopropyltriethoxysilane, (3,3,3-tri¯uoropropyl)-trimethoxysilane)

Silanes in the form of RSiX3 (where X represents a halide, alkoxide, or alkyl group; and R an organofunctionality) are often referred to as `coupling agents' because they can act as a bridge between organic and inorganic layers. X is a hydrolyzable, leaving group that attaches the Si to the surface, and the functional groups R are chosen to form protective layers, chemically modify electrodes, or immobilize large functional groups such as biomolecules on surface [655]. A series of silanes on clean and water-predosed TiO2(1 1 0) surfaces were investigated byCampbell and co-workers [630,655± 657], see Table 16(f). The simplest molecule, TEOS, consists of a central Si atom bound to four ethoxygroups (EtO, where

Et equals C2H5). At low coverages it readilydissociates, splitting off one or several EtO groups. Multilayers form at high dosages at low temperatures. Heating of a TEOS monolayer releases ethanol and ethylene in a 1:1 ratio around 650 K, suggesting a transfer of hydrogen between ethoxy species as in

2EtOads ! EtOHgas C2H4;gas Olattice (34) 200 U. Diebold / Surface Science Reports 48 (2003) 53±229

This reaction leaves a lattice oxygen at the surface, similar to the reaction of ethoxys on TiO2(0 0 1) [629]. The presence of pre-dosed water facilitates the dissociation of the TEOS molecule, probablyvia elimination of the EtO groups via reaction with water. It also leads to a low-temperature (350 K) desorption peak of EtOH. Annealing adsorbed TEOS to 750 K (with or without co-adsorbed D2O), produces a disordered SiO2 layer in the net reaction

Si OEt4;ads ! 2EtOHgas 2C2H4;gas SiO2;ads (35)

Si OEt4;ads D2Oads ! 2EtODgas EtOHgas C2H4;gas SiO2;ads (36)

Vinyltriethoxysilane (VTES, where one of the Et groups in TEOS is substituted by a ±CH=CH2 group, see Table 16(f )) and diethyldiethoxysilane (DEOS, basically a TEOS molecule but with two EtO groups substituted byEt) also dissociate on the clean TiO 2(1 1 0) surface in the absence of water by splitting off two EtO groups [655]. High-temperature reaction again produces a SiO2 surface layer with ethanol, ethylene, and (in the case of VTES) CHBCH as reaction products. Aminopropyltriethoxysilane

(APS, with a (CH2)3±NH2 group substituting one EtO in TEOS) does not dissociate on TiO2(1 1 0) in anymeasurable quantity [655]. The fourth silane investigated byCampbell and co-workers [657], (3,3,3-tri¯uoropropyl)trimethy- loxysilane (FPTS, where Si is coordinated to three ±OCH3 groups and a ±CH2±CH2±CF3 group) turned out to be more reactive than the ethoxysilanes considered. This molecule also adsorbs dissociatively bysplitting off ±OCH 3 groups. SSIMS showed that these bind to exposed Ti sites, and that the Si in the remaining (CF3CH2CH2)Si(OCH3) complex binds to two lattice oxygens (two bridging oxygen atoms in the proposed model [657]). The ±OCH3 groups decompose at 550±650 K to form methane, formaldehyde, and methanol gases. The CF3CH2CH2 ligand also decomposes around 620 K. Water facilitates the formation of methanol which desorbs at considerablylower temperatures ( 300 K).

5.3. Photocatalysis on TiO2

The photoactivityof TiO 2 is one of its technologicallymost attractive properties. The creation of electron±hole pairs through irradiation of light, either in TiO2 itself, or in adsorbed molecules, and the following chemical or electron transfer reactions are at the heart of TiO2-based photodevices applied in a range of areas. Intense research was initiated byFujishima and Honda's [25] discoverythat water can be photocatalytically split into hydrogen and oxygen. As this discovery was made at the height of the oil price hikes in the seventies, it was immediatelyembraced as an inexpensive and safe wayto produce energy.However, the use of TiO 2 for hydrogen production never reached commercialization, although there is a recent interest in reviving this issue. Manyother useful reactions can be induced over irradiated TiO2, and a verybrief review is given in the next section. The band gap of TiO2 is relativelywide and its absorption properties are not well matched to the frequency spectrum of natural sun light. The functionalization of the TiO2 surface with dye molecules has made TiO2-based solar cells much more ef®cient for the conversion of solar energyinto electric energy (Section 5.3.2). Manyfundamental questions on understanding and improving the photocatalytic activityof TiO 2 are open at this point, and maybe addressed bystudyingphotocatalytic reactions on single-crystalline surfaces under UHV conditions. The current literature on such studies is summarized in Section 5.3.3. U. Diebold / Surface Science Reports 48 (2003) 53±229 201

Fig. 70. Band structure in an n-type semiconductor (a) before contact with an electrolyte (¯at band situation) and (b) in contact with an electrolyte.

5.3.1. Heterogeneous photocatalysis

Various aspects of the photocatalytic activity of TiO2 are reviewed in [30,38,41,43,658]. The reactions accomplished through photocatalysis can involve oxidations and oxidative cleavages, reductions, geometric and valence isomerizations, substitutions, condensations, and polymerizations

[658]. Of special interest is the employment of TiO2-based photocatalysts in remediation processes. Here complete mineralization is desired, i.e., the conversion of organic compounds to CO2,H2O, À NO3 , or other simple basic products. The primaryexcitation process results in an electron in the conduction band and a hole in the valence band (Fig. 70a, Eq. (37)). When TiO2 is in contact with an electrolyte, the Fermi level equilibrates with the redox potential of the redox couple. The resulting Schottkybarrier ( Fig. 70b) drives the electron and the hole in different directions. (It has been pointed out that the picture of a Schottkybarrier formation is no longer valid when size of the photocatalytic TiO2 particles approach nanoscopic dimensions [659].) The components of the electron±hole pair, when transferred across the interface, are capable of reducing and oxidizing an adsorbate, forming a singlyoxidized electron donor and a singlyreduced electron acceptor, Eqs. (37)±(42).

À TiO2 hn ! TiO2 e ; h (37) TiO2 h RXads ! TiO2 RXads (38) TiO2 h H2Oads ! TiO2 OHads H (39) À TiO2 h OHads ! TiO2 OHads (40) À À TiO2 e O2;ads ! TiO2 O2 (41) À À TiO2 e H2O2;ads ! TiO2 OH OHads (42) These processes result in anion or cation radicals which can undergo further reactions. Hydroxyl radicals are generallyconsidered the most important species in the photocatalyticdegradation of organics, although not in UHV-based studies (see Section 5.3.3). It is generallyheld that hole capture is directlythrough OH and not via water ®rst, i.e., through Eq. (40) rather than Eq. (39). The 1b1 orbital of water lies above the 1p level of OH [477], so one might expect water to be better at `capturing' a hole than OH, but the radical-cation of water maybe neutralized before decomposing into an OH radical. 202 U. Diebold / Surface Science Reports 48 (2003) 53±229

Also, it is mostlyassumed that the surface is OH covered and therefore the hole is directlytransferred to OH. UHV studies would be needed to correct or reinforce this assumption.

It has been observed that the photocatalytic activity of TiO2 is completelysuppressed in the absence of an electron scavenger such as molecular oxygen. Because the conduction band of TiO2 is almost completely isoenergetic with the reduction potential of oxygen in inert solvents, adsorbed oxygen À serves as an ef®cient trap for photogenerated electrons. The resulting species, superoxide, O2 ,is highlyactive and can attack other adsorbed molecules. Several other oxidation processes, in addition to reactions (Eqs. (38)±(42)) can occur as well [38,41,500]. Often, loading of TiO2 with Pt and addition of H2O2 (Eq. (42)) enhance the overall ef®ciencyof the photocatalyticdegradation processes. In order for photocatalysis to be ef®cient, electron±hole pair recombination must be suppressed before the trapping reactions occur at the interface. The recombination reaction occurs veryfast, and the resulting low quantum ef®ciencyis one of the main impediments for the use of TiO 2. While degradation of airborne pollutants has lead to an explosion of TiO2-permeated paints and papers to clean up everything from cigarette smoke to acetaldehyde, TiO2-based systems to treat contaminated water are severelysuffering from the low yield, and their economic feasibilitywas questioned [660].

5.3.2. Photovoltaic cells

The use of TiO2-based systems for the conversion of sunlight into electric energy involves a range of interesting fundamental questions. The most successful device is the so-called `GraÈtzel cell'. Fig. 71a shows the functioning principle of this dye-sensitized photovoltaic cell, which was originally proposed in [29]. The fundamental and technical aspects of this cell are reviewed in [659,661]. It consists of an electrode covered with colloidal, nanostructured TiO2, which has been sintered together to form a percolating network. The mesoporous structure of the ®lm exhibits a porosityof up to 50% and provides a veryhigh surface area. The pores between the particles are interconnected. The colloidal

TiO2 ®lm is supported on glass that is covered with a transparent conducting oxide (TCO) layer. The surface of the colloidal TiO2 ®lm is covered with a monolayer of dye. Photon excitation of the dye (S) results in an electron in an excited state (SÃ). This electron is injected veryrapidly(within a few picoseconds) into the conduction band of TiO2. The backward reaction, i.e., charge recombination, occurs at a much slower rate (within a millisecond time frame). The electron injected in the TiO2 conduction band then travels through the colloidal ®lm to the back electrode. Interestingly, the as- prepared nanostructured TiO2 ®lm has a veryhigh electrical resistivitybut conducts electricityquite well when illuminated [659,661]. The transport through the mesoporous ®lm is not well understood, possiblytrap states within the particles or on their surfaces, as well as screening of the electron bythe electrolyte, are major factors. The electrode with the TiO2 ®lm is connected to a counterelectrode (a piece of glass, covered with a TCO layer). The gap between the electrodes is typically ®lled with a À À À molten salt containing a redox mediator (A/A ). The iodide-triiodide redox couple (I =I3 ) works best, and the redox reaction is catalyzed by a small amount of Pt on the surface of the counterelectrode. The electric circuit is closed when the photo-oxidized dye molecules are reduced by electron transfer from the electrolyte. The operating voltage of the cell is given by the potential difference between the conduction band of the colloidal TiO2 and the redox potential (relative to an SCE electrode) of the À À I =I3 redox couple. In the case of the GraÈtzel cell, the voltage is in the order of 0.4 V, and currents between 16 and 20 mA/cm2 can be achieved under standard operating conditions. Currentlythe overall ef®ciencyof the photovolatic cell is in the order of 10%. U. Diebold / Surface Science Reports 48 (2003) 53±229 203

Fig. 71. (a) Schematic representation of the principle of the dye-sensitized photovoltaic cell [29]. Indicated are the electron energylevels in the three different phases important for the functioning of the cell. The systemconsists of a RuL 2(SCN)2 dye À À adsorbed on a colloidal TiO2 layer and the I /I3 redox couple in the electrolyte. The cell voltage under illumination corresponds to the potential difference, DV, between the quasi-Fermi level of TiO2 and the electrochemical potential of the electrolyte. S stands for sensitizer (the dye); SÃ, electronicallyexcited sensitizer, S , oxidized sensitizer. (b) Molecular orbital diagram for ruthenium complexes anchored to the TiO2 surface by a carboxylated bipyridyl ligand. The visible light absorption of these types of complexes is the so-called metal-to-ligand charge transfer (MLCT) transition. The carboxylate groups are directlycoordinated to the surface titanium ions producing intimate electronic contact between the sensitizer and the semiconductor. From Hagfeldt and GraÈtzel [659]. # 1995 The American Chemical Society. 204 U. Diebold / Surface Science Reports 48 (2003) 53±229

The interface between the sensitizer and the TiO2 surface is critical for the functioning of the cell and is of fundamental interest for surface investigations [615,662,663]. Several sensitizers were tried, and 0 0 the most promising ones have the general structure ML2(X)2, where L stands for 2,2 -bipyridyl-4,4 - dicarboxylic acid, M for Ru or Os, and X for halides, thiocyanate, or water. The structure of the most often used dye, the ruthenium complex cis-RuL2±(NCS)2, is sketched in Fig. 71b. This dye, often called `N3', is verystable and accomplishes a close to quantitative photon to electron conversion in the spectrum of visible light. (Recently, a black dye was found that shows a better light absorption behavior in the infrared region [661].) The photoexcitation process occurs between a ground state electron on the Ru center to a ligand orbital (metal to ligand charge transfer, MLCT, see Fig. 71b). The dye is anchored with two carboxylate groups to the TiO2 surface, possiblyin a fashion discussed in Section 5.2.1.9 and Fig. 67. The fast electron injection, which is essential for the functioning of the cell, is attributed to the hybridization between the pÃ orbital of the carboxylate with the 3d electrons of the exposed ®vefold coordinated Ti ions [663]. The backward reaction is slow because it involves electron transfer from

TiO2 to the oxidized ruthenium metal whose electronic overlap with the Ti wave functions is small, see Fig. 71b. As a model system, the adsorption of bi-isonicotinic acid on rutile (1 1 0) was studied with

XPS and NEXAFS [615], see Section 5.2.1 and Fig. 67. The colloidal TiO2 ®lm has the anatase structure, however, and mostly(1 0 1), (1 0 0), and (0 0 1) faces are exposed. The adsorption of formic acid on anatase (1 0 1) was modeled with ®rst-principles calculations [605]. Electrochemical measurements on anatase (1 0 1) single crystals show a much smaller ef®ciency for electron injection from the dye to the substrate as compared to nanophase ®lms (with mostly the (1 0 1) face exposed) [209]. Perhaps this is a sign for the importance of surface defects in the injection process.

5.3.3. Photocatalysis on single-crystalline TiO2

5.3.3.1. Oxygen, water, CO, and CO2. In UHV-based photocatalytic studies, vacancies in the bridging oxygen rows of TiO2(1 1 0) playa major role [43,444,496±498,500,522,531,664±668]. As discussed above (Section 5.1.3), these vacancies serve as adsorption site for O2 molecules. Two species, originallytermed à'and`b' state byYates and co-workers [496,497,522], were observed to chemisorb at vacancies. The a species, produced byadsorption at 105 K, can be activated to oxidize CO [496,497,522]. Lattice oxygen is not chemicallyinvolved in the CO 2 production [522], and, interestingly, the reaction is not enhanced by the presence of Pt [444]. Depending on surface preparation, two different defect states, termed à1' and à2' state, can be created on the TiO2(1 1 0) surface [498], which are both active for CO2 production. Dosing oxygen at elevated temperatures, or annealing a sample with oxygen in the à' state, produces the `b' state which photodesorbs and does not catalyze the CO oxidation reaction. Oxygen vacancies and molecular oxygen adsorbed at these defect sites also proved important in the photo-oxidation of CH3Cl/TiO2(110) [664,665]. The threshold for photoexcitation is 3.1 eV, indicating that excitations within the TiO2 substrate are important. Interestingly, surface hydroxyl groups do not oxidize methylchloride in the absence of adsorbed oxygen, which points out that Eq. (41) is more important than reactions (39) and (40), at least under UHV conditions. The ternary system consisting of molecular oxygen, water, and rutile (1 1 0) has been investigated with TPD and spectroscopic techniques [500]. Molecular oxygen was dosed on a TiO2(1 1 0) surface with oxygen vacancies at 115 K. Irradiation with 4.1 eV photons lead to the photodesorption of the O2 molecules adsorbed at the oxygen vacancies. However, the adsorbed oxygen is not removed completely and a photocatalytically active intermediate is formed in reaction with water. This intermediate was U. Diebold / Surface Science Reports 48 (2003) 53±229 205 tentativelyassigned as hydrogenperoxyspecies. When the oxygen-precovered surface was covered with thick ice layers and irradiated with photons, a very sharp O2 peak was observed in TPD spectra. This was interpreted as a trapping of the O2 molecules in the ice layers.

5.3.3.2. Alcohols. Brinkleyand Engel [666±668] studied the photocatalytic reaction of 2-propanol to acetone and water on TiO2(1 1 0) and TiO2(1 0 0). Onlythe most tightlybound propanol molecules were reactive for the photocatalytic dehydrogenation reaction, and no reaction was found in the absence of co- adsorbed O2. Electron trapping at the adsorbed O2 molecule (Eq. (41)) was invoked. From a combination of TPD and molecular beam experiments, it was found that about 0.08% of surface sites are active. The presence of oxygen vacancies increases the number to 0.15. It was suggested that steps, kinks, oxygen vacancies, and other sites that can adsorb O2 stronglyare active sites for the photocatalytic reaction. An intrinsic `heterogeneity' of the `fully oxidized' surface was concluded from these studies. This observation is in line with the quite complex nature of oxidized TiO2(1 1 0) surface revealed in more recent STM studies of oxidized surfaces (see Section 2.2.2). Interestingly, the oxidation of 2-propanol is dominated by a thermal reaction channel on the TiO2(1 0 0) surface. The thermallyactivated reaction on this surface is attributed to the proximityof bridging oxygen atoms to the propanol adsorption sites (the exposed Ti(5) atoms), which leads to a thermallyactivated dissociation. Henderson [174] suggested a similar mechanism to account for the difference in the water dissociation reaction on these two surfaces, see Section 5.1.2. Recently, embedded cluster calculations of an adsorbed water molecule on a small rutile (1 1 0) cluster were performed [669]. It was found that the excited electron is located in the subsurface region, and that the hole is localized at the surface oxygen atoms and enhances the radical character of the hydroxyl group, in agreement with the notion that the OH radical is the active species for TiO2 photocatalysis (Eq. (40)).

5.3.3.3. CHX3 (X Cl, Br, I). A series of adsorption and photodesorption studies was performed by Stair, Weitz, and co-workers [670±678], see Table 16(g) Methyl iodide (CH3I) weaklyphysisorbsat TiO2(1 1 0) and the monolayer desorbs around 170 K [674]. The morphologyof CH 3I films was deduced from TPD and UV-light desorption measurements [678]. At 90 K, films grow stochastically. Depending on the annealing temperature, either a Stranski±Krastanov type morphology (2D film with 3D clusters after annealing to 120 K) or a 2D film can be created. From TOF-REMPI measurements it was concluded that CD3I molecules arrange themselves in a flat orientation in the monolayer regime [677]. The C±I bond lies roughlyparallel to the surface normal, and the molecules are in an antiparallel configuration in the multilayer regime.

Laser irradiation in the UV regime produces CH4,C2H6,I2,C2H2,C2H5I2,C2H5I [674,675]. The laser-induced desorption of CH3I ®lms was studied in dependence of the surface coverage. The initial excitation involves moving an electron from the TiO2 substrate into a vertical af®nitylevel [671,676,677]. Presumably, CH3I molecules in the monolayer regime desorb in an Antoniewicz-type process (where the initial excitation of the adsorbed molecule results in an ion which moves towards the surface and is expelled into the vacuum when it is neutralized [679]). When the coverage is increased, the signal of desorbing CH3I molecules is weakened due to collisional interactions within the ®lm. For thicker ®lms (>5 ML), the photodesorption signal increases again. It was proposed that photoexcited electrons solvate in the overlayer, which then induces desorption from the CH3I ®lm's surface. The photodesorption of CH3Br behaves similarlyto CH 3I. In contrast, CH3Cl was not found to photodesorb, when adsorbed byitself. Photodesorption was observed when it was co-adsorbed with CH 3I. 206 U. Diebold / Surface Science Reports 48 (2003) 53±229

6. Summaryand outlook

Titanium dioxide is a fascinating material from a surface science point of view. So much is known now about its surface geometric and electronic structure, and about fundamental steps in surface reactions, yet, so much is still to be learned. The main lessons that can be drawn from the previous chapters are summarized in the following paragraphs. In addition to gaining an ever better understanding of surface structure, geometry, and reactivity of TiO2 itself, two issues will be important in future research. The ®rst is, how can the lessons learned from TiO2 be applied to other metal oxide systems? The surface science of metal oxides is a still growing ®eld, and it will be interesting to see the applicabilityand limits of using TiO 2 as a prototype for other metal oxides. The second issue is probablyeven more important, how can the fundamental knowledge at the molecular scale, obtained in surface studies, be put to direct use for the manyapplications where TiO 2 surfaces playa role? As outlined in Section 1.2 of this paper, TiO2 is a technologicallyveryimportant material, and the insights from surface-science studies could possiblyhelp in a wide range of technical areas useful to society.A few thoughts related to these emerging topics are given in the following.

6.1. What has been learned and what is missing?

TiO2(1 1 0) is now well accepted as the quintessential bulk model oxide and this surface is being used routinelyas a starting point for exploring new phenomena on oxides. Yet, the systemis complex enough that it might still deliver some surprises. The combination of spectroscopic, diffraction, and imaging techniques, applied bymanydifferent groups world-wide over the course of manyyears,has delivered a verycomplete picture of the rutile (1 1 0) surface. The stoichiometric surface and manyof the different con®gurations of defects are now well characterized. Simple schemes, such as the concept of autocompensation (see Section 2.2.1.1), are a veryuseful ®rst approach to interpret experimental results. Yet, one has to be careful; too simplistic approaches (e.g. explanations for reconstructions that involve an ordered arrayof oxygenvacancies or microfacets) have proven wrong upon closer inspection. TiO2 has also become a model system for theorists to test, sharpen, and re®ne ®rst- principles calculations of complex materials. The combined effort of experiment and theoryhas lead to a level of knowledge that is certainlyunsurpassed for anymetal oxide. Still, it is somewhat exasperating that important discrepancies between theoryand experiment are still unresolved, such as the relaxation of the bridging oxygens (Section 2.2.1.2), the position of the defect state in the band gap on reduced surfaces (Section 3.2.1), and the adsorption state of water on rutile (1 1 0) (Section 5.1.2).

From an experimental point of view, TiO2(1 1 0) is a wonderful sample to work with, but the exact preparation conditions are veryimportant for the surface geometric and defect structure ( Section

2.2.3)Ðwhich can be curse and opportunityat the same time. Other orientations of rutile TiO 2 are still less well understood; most notablythe TiO 2(0 0 1) surface, where a con®rmed atomic model of common reconstructions is still lacking. This is particularlyunfortunate, as it has been the substrate of choice for manyreactions of more complex organic molecules, see Section 5.2, and of many

(photo)electrochemical studies on single-crystalline TiO2 [471,680±695]. Surface chemistrystudies of organics are particularlyimportant as theyprovide a link between fundamental research and applications in heterogeneous catalysis. It would be useful if relatively complex reactions were performed on the much better characterized TiO2(1 1 0) surface, and if the atomic-scale mechanisms were re-inspected, once a more accurate picture of the atomic geometryhas been worked out. U. Diebold / Surface Science Reports 48 (2003) 53±229 207

The `other' TiO2 structure, anatase, which is technologicallyquite relevant, is now being investigated (Section 2.6). One will have to ®nd out if the use of thin ®lm and mineral samples is practical, but the ®rst results on this material, and the progress that is being made, is quite promising.

The growth of metal overlayers, clusters and ultrathin ®lms on TiO2 has yielded a very complete picture. Anyresearcher interested in metal/oxide systemscan learn much from the material summarized in Section 4. It is beautiful to see how the main traits of metal overlayers follow general trends across the whole periodic table, and how this ties in with the rich details available for almost any metallic adsorbate on TiO2. The wealth of results on metals/TiO2 should give scientists, especially theorists, a good playing ®eld for pushing our understanding of the technologically so important metal/ metal oxide interface. The next step in the attempt to provide more insight into heterogeneous catalysis should certainly involve chemistryon these metal/oxide systems. Will the metal/TiO 2(1 1 0) system be a useful model in this respect? The characterization of encapsulated Pt and Pd clusters in the SMSI state (see Sections 4.2.24 and 4.2.25 on the growth of Pd and Pt) seems to have happened after the surge of interest in these systems is over, but the issue of gold-related catalysis is currently quite `hot' and the surface science of TiO2, especiallyin relation to defect structures, might help to come up with good answers to manyof the open questions ( Section 4.2.28). Perhaps one of the main lessons from recent research is the close relationship between bulk and surface properties in TiO2. It has been known for a long time that surface defects are important on metal oxides, but now it is clear that bulk defects have to be considered as well. The equilibration between bulk and surface defects that takes place at somewhat elevated temperatures has serious consequences for the surface morphologyand structure ( Section 2.2.2) as well as surface chemistry( Section 5.1). An adsorbate is affected bybulk defects provided it: (1) interacts stronger with surface defects than with the perfect surface (which, in essence, means almost any molecule or atom), and (2) the interaction with the surface takes place under conditions where the rate of diffusion of subsurface defects to the surface is comparable to the adsorption rate (i.e., at onlyslightlyelevated temperatures or pressures). In this sense, the bulk defect concentration is quite an important property. It has become a common practice among researchers to quote the crystal color (see Fig. 5) as a qualitative measure for the defect concentration, but this is certainlyan area where more quantitative studies about the different kind of defects, their depth distribution and diffusion properties towards and across the surface are needed.

6.2. TiO2 in relation to other transition metal oxides

Considering at the mere number of publications (Fig. 1), TiO2 is the model transition metal oxide. Yet, how much of the results and concepts surveyed in the proceeding chapters can be and will be applied to the surface science of other metal oxides? One point that cannot be emphasized stronglyenough is to `look before youdo'Ð ®rst perform a thorough characterization of the surface structure and morphology before performing other experiments. After all, much of the results discussed in Section 2 have onlybeen found recently. Who would have anticipated that the prevalence of `(1 Â 2)'-type rows, even when LEED shows a (1 Â 1) surface, the `rosette' structure, or the in¯uence of the bulk state on the surface morphology

[156]? For historic reasons, much of TiO2-based research was performed before STM and AFM with (near) atomic resolution was applied. What is now known about the local surface structure puts in question parts of this results; something that can and needs to be avoided with other oxide system. As 208 U. Diebold / Surface Science Reports 48 (2003) 53±229

Henrich and Cox [1] have pointed out in their 1993 monograph, surface preparation of metal oxides is a research project in itselfÐthe experience with TiO2 certainlyproves this statement. Other oxides with rutile structure. For the class of metal oxides with the rutile structureÐSnO2, RuO2, CrO2, MnO2,VO2Ðexploring the similarities and also differences to rutile TiO2 should be valuable in two respects. First, progress on these other oxides will proceed much more rapidly, without manyof the detours taken in TiO 2. Second, it will be interesting to see how manyof the structural and chemical properties are speci®c to the rutile structure and what should be attributed to the Ti cation.

For example, the second-most studied oxide surface with rutile structure, SnO2, is still relatively unexplored. Upon annealing a SnO2(1 1 0) surface, a series of reconstructions evolve that have been characterized with LEED [696]. More recently, some of these reconstructions have been observed with STM [697±699] (as pointed out, a necessary®rst step to really understand this surface). For these structures, ordered arrays of oxygen vacancies have been suggested [696,700,701]; an approach that has fabulouslyfailed for TiO 2 (Section 2.2.2). Of course one should not jump to conclusions too quickly and rule out that ordered O vacancies indeed do form on SnO2Ðit could well be possible that the fact that Sn is most stable in oxidation states 4 and 2 stabilizes these structures, much in contrast to Ti, which has a rich phase diagram with manystable phases ( Fig. 4). Still, a look at TiO2 suggests to test other possibilities, for example, structures involving interstitial Sn atoms. Metallic ruthenium and its interaction with oxygen has long been investigated for its use as a possible catalyst for CO oxidation [702]. Perhaps not unsurprisingly, ruthenium is oxidized under reaction conditions, and the resulting RuO2(1 1 0) ®lm is the catalytically active phase [703]. This observation has stimulated a ¯urryof activityand has caused RuO 2 to become a runner-up (with SnO2) to be the second-most investigated oxide with rutile structure. Vacancies in the rows of bridging oxygens are an issue here as well. However, quite central for the activityof RuO 2(1 1 0) are oxygen atoms adsorbed at `cus' sites plays, i.e., in on-top sites at ®vefold coordinated Ru atoms [504,704]. Even under UHV conditions such sites are occupied at slightlyelevated pressures. No indication for these oxygensat

TiO2(1 1 0) so far; could such sites become important when one goes to high even higher pressures? Because RuO2 is a metallic material, imaging with STM is possible over a much wider range of bias voltages. Bright protrusions in STM images are attributed to O atoms [705], in contrast to STM on

TiO2(1 1 0) (Section 2.2.1.3)Ðmaybe a warning sign that `imaging the surface cation' is not automaticallya given in atomicallyresolved STM of metal oxides. So far almost all experiments have been performed on the thin ®lm that forms on Ru upon oxidation. RuO2 single crystals of suf®cient size can be grown [706], but have not yet been widely employed in surface studies. As pointed out above, the interaction with the bulk is important for the surface chemistryof TiO 2Ðperhaps this is something to consider also in the case of RuO2. For the other isostructural oxides, MnO2,VO2, and CrO2, large single crystals are not available, but rutile TiO2(1 1 0) acts an important substrate in growth experiments, see Section 4.2 and Table 6.These oxides exhibit quite interesting bulk and/or surface properties, e.g. the half-magnetic ferromagnetic nature of CrO2 or the catalytic activity of TiO2-supported VO2. As has been mentioned above, the growth of metal oxide on TiO2 is a somewhat neglected ®eld, at least compared to the growth of metal overlayers on this substrate. The possibilityof a controlled and tailored growth of these interesting oxides on

TiO2(1 1 0) should provide impetus to look more closelyinto interface issues of oxide/oxide systems. Other titanates. CertainlyTiO 2 is of importance for other titanatesÐthe (0 0 1) surface of perovskite titanates such as SrTiO3 or BaTiO3 can be terminated with a surface layer of either TiO2 or MO (M Sr, B, Ca, etc.) stoichiometry. Because perovskite oxides are widely used as substrates for U. Diebold / Surface Science Reports 48 (2003) 53±229 209 epitaxial growth of superconductors and dielectric materials, there is some interest in exploring the best recipes for surface preparation of SrTiO3 [67,707]. For example, it is well-known that an etching procedure produces an exclusivelyTiO 2-terminated surface [707]. Interestingly, the reconstructions that occur upon high-temperature annealing of SrTiO3 are not yet completely understood. Again, in the light of TiO2-based research, it is questionable if ordered oxygen vacancies (which seem to come to mind immediately) are a good model, and recent studies indicate that this indeed not the case [708,709].On the other hand, point defects, a signi®cant issue in TiO2 surface chemistry, might be both similar and different because of the presence of another cation. Other oxides. Onlya few semiconducting oxides can be investigated in bulk form with surface science techniques. Often the alternative approach is taken, i.e., growing a crystalline ultrathin ®lm on a metallic substrate [3,287,710]. It is often a matter of necessityto use these systems,especiallywhen oxides are too insulating as bulk single crystals. Again, one important lesson from TiO2 is that the bulk does matterÐfor the formation of surface defects, during surface preparation, for adsorption processes, in reactivityor encapsulation experiments. Maybethis is not so important in the case of non-reducible oxides such as MgO or Al2O3, but for reducible transition metals oxides that exhibit several phases, it is likelythat similar processes happen at relativelylow temperatures. In this respect, it might be interesting to compare the reactivityof ultrathin oxide ®lms (maybeeven an ultrathin TiO 2 ®lm) one-to- one with studies on bulk-crystals in order to ®nd out just what the applicability and limitation of ®lm model systems are.

6.3. TiO2Ðmixed and doped

Most metal oxides, when used in the technical applications are `functionalized' through doping, or are used in combination with other oxides, and so is TiO2. For example, doping with traces of earlytransition metals and/or lanthanides increases the quantum yield in photocatalytic degradation processes, and TiO2 in gas sensors is usuallydoped with noble metals. Indeed, most of the other industrial applications cited in

Section 1.2 employdoped TiO 2 instead of the pure material. Doping can have a varietyof consequences as it mayalter the structure, electronic properties, or thermal stabilityof this material. From a surface science perspective, an unraveling of all these factors is an almost completelyunexplored (though possiblyhighlyrewarding) territory.There is now suf®cient information on pure TiO 2 that one can con®dentlygo one step further and introduceÐin a well-controlled wayÐimpurities and dopants. This brings surface studies not onlyone step closer to complex reality,the few examples of surface investigations on doped TiO2 [153,374,376,711,712] show that the local variations of electronic structure are also quite interesting from a fundamental point of view. The most striking example is the in¯uence of

Co doping on the magnetic properties of TiO2 anatase [55,56]. This novel dilute ferromagnetic semiconductor was found in a combinatorial studybyMatsumoto et al. [55]. While doping has long been used to improve the performance of TiO2-based devices, rationallytailoring the surface properties of this material with atomic-sized control should provide opportunityto create speci®c surface structures, control defects and their diffusivity, and tailor electronic properties.

6.4. Nanostructured TiO2

The current international trend (or, more accurately, `hype') to synthesize, characterize, and investigate `nano'materials has also embraced TiO2. This material lends itself quite well to building 210 U. Diebold / Surface Science Reports 48 (2003) 53±229 tinystructures in all sorts of sizes and shapes. Such nano-TiO 2 is typically produced in a sol±gel process, where a titanium alkoxide or halide (TiCl4,TiF4) is hydrolyzed, often in the presence of a template such as nano-spheres, nano-rods or anodic porous alumina [713,714,715], but other techniques have been used as well [716±719]. The structures that have been madeÐnano-rods, whiskers, wires, spheres, ordered holesÐare simplyfascinating. For example perfectlyordered TiO 2 `nano-whiskers', i.e., ¯at anatase platelets a few nanometer wide and several tens of nanometer long, were produced byZhu and Ding [720].In this case, the selective coordination of acetate groups on speci®c lattice planes was invoked to account for the oriented growth. (Incidentally, the speculation about the anchoring of this group to the different surfaces in [720] could certainlybe answered bysurface scientists, considering how much is now known about acetates on rutile, Section 5.2.1. Just one example how interrelated

`nano' and surface research are.) Another beautiful construction are hollow TiO2 `microspheres', ca. 20 mm in diameter with a wall onlya few tens of nanometer thick, synthesized byIida et al. [721]. Yin et al. [722] have succeeded in `functionalizing' such hollow spheres bypacking small Ag particles onto the interior surface of the inside void. It was suggested that these nanoparticles could then be etched out, leaving behind nanoholes in microspheres. It was speculated that these structures could be used as extremelytinycontainers for encapsulation in the deliveryof drugs or the protection of biologicallyactive agents [722]. Or one could dream as these spheres becoming the tiniest `microreactors' for catalytic reactions. More mundane applications would be as a low-density ®ller materials.

The formation of periodic arrangements of nanostructured TiO2 can also be achieved. For example, the formation of a zeolite-like mesoporous materials has been accomplished for TiO2 and other metal oxides [719,723]. Such high-surface area materials could be interesting for catalytic applications or in photonics. The formation of ®lms with a self-organized nanostructure is explored for improving devices based on photo-active TiO2 (Section 5.3.2). A hexagonal arrangement of anatase nanocrystals was achieved [724] and rod-like single-crystalline anatase particles could be brought into a regular cubic array [715]. These periodic arrangements are achieved through a balance of electrostatic forces in the solution. Theyexhibit an extremelyregular pore size and a high surface area. The smallest nano-rods in [715] have predominantly(1 0 0)-oriented surfaces (a surface that has just recentlybeen characterized with surface science techniques [231]). Such a structure would allow to investigate the in¯uence of surface orientation on the performance of dye-sensitized solar cells (Section 5.3.2) or electrochromic devices.

Interestingly, TiO2 nanostructures are almost invariablyeither amorphous or of anatase formÐ another good motivation for surface scientists to ®nd out more about the surfaces of this TiO2 polymorph. There is a clear connection between the surface properties, the rational development of improved synthesis routes, and the possible usefulness in applications of nanomaterials. Hence the lessons, learned from atomic-scale investigations such as the ones reviewed in this paper, could be quite valuable for synthesis-oriented researchers. Conversely, using well-characterized nanostructures instead of single crystals in surface chemistry experiments might be a promising and exciting new approach. Research `in the nanoworld' provides almost unlimited nourishment for imagination on how these tiny structures could be usedÐand often a glaring gap to real technical developments. In the case of TiO2, however, the proven performance improvement of nano-scaled structures in photo-active devices shows that a cross-fertilization between surface studies and synthesis of nano-TiO2 could be valuable well beyond the realm of fundamental research. U. Diebold / Surface Science Reports 48 (2003) 53±229 211

6.5. Going beyond single crystal and UHV studies

If research on oxides is viewed as an important attempt to bridging the (in)famous materials gap, then research on `dirty' oxides (in the sense described above, i.e., mixed, doped, nano-sized, polycrystalline, etc.) will certainly constitute the next logical step. Of course the key factor is doing so without sacri®cing the degree of control that is provided byUHV studies and single crystals. In this sense, newlyemerging techniques are particularlyimportant, such as the SEM/STM that has been applied byAsari and Souda [725] to image selected corns of polycrystalline Degussa P25 powder. More of such studies are needed, especiallyin combination with the tailored nanostructures mentioned in Section 6.4. If one wants to link surface-science studies closer to applications, it is also vital to bring surface science out of UHV (`bridging the pressure gap' as it is often termed [726]). This would further test the relevance of the acquired knowledge for technical applications. Manyspeci®c questions are awaiting an answer, for example, How stable are the various defects under high-pressure environments? Are point defects as important for adsorption and reactions as UHV-based adsorption experiments suggest? Do step edges (which do not seem to be too critical under UHV conditions) playa more important role in a high-pressure environment? Can one link surface reactivityand local coordination, i.e., bypurposely creating arrays of undersaturated atoms (e.g. in a rosette network)? Is the interaction with the bulk alreadya signi®cant factor at room temperature if one goes to high enough pressures? Scanning probe, spectroscopic and optical techniques, compatible with high pressures, are currentlybeing developed

[726±730].TiO2 could be a wonderful test system for these newly emerging techniques if one considers its often defect-driven surface chemistry(and, otherwise, relative inertness) combined with clear spectroscopic evidence for defect structures.

As pointed out several times throughout this review, the photoactivityof TiO 2 is one of its most interesting and attractive property.Studies of photoactivitycan convenientlybe combined with UHV techniques, as shown in an impressive waybyYates and co-workers. A clear outcome of these studies (see Section 5.3.3) is the dominating role of defects. Probing into the effect of the local environment on photoactivity, e.g. with low-T STM, is an experiment that just waits to be taken up bysomeone. UHV-based photocatalytic studies have shown some important differences to chemistry in a wet environment, e.g. the (relative) unimportance of hydroxyl groups [664] or the absence of the enhancing effect of Pt for CO oxidation [444]. The combination of (photo-) electrochemical measurements with surface analytical techniques, e.g. in reaction cells attached to UHV chambers, quite successful for metal surfaces, has hardlybeen explored for TiO 2. There are ample examples of photoelectrochemical measurements on TiO2 single crystals [28,209,227,471,680±695,731]. Unfortunately(from a surface science perspective), most of these studies have been performed on the rutile (0 0 1) surface, which is an inherentlyunstable surface with an atomic-scale structure that is not yet resolved (Section 2.4). Also, from today's point of view, most of these surfaces were poorlycharacterized; the usuallyapplied preparation techniques of reduction in an H 2 atmosphere and polishing and etching can lead to segregation of impurities and will introduce manysurface defects [695,732]. Still, there are manyindications that surface defects playa big role in the photochemical activity [209,680,731]Ðit is just not known which ones and how. With the background of atomic-scale investigations of surface defects (Section 2.2.1.4), it would be well- worth an effort to ®nd out how different surface terminations, orientations, and structures in¯uence photo-reactivity. 212 U. Diebold / Surface Science Reports 48 (2003) 53±229

6.6. Concluding remarks

As is often the case when reviewing the current literature on a subject, glaring gaps and omissions become clear. While much work has been done, and much has been achieved, the surface(s) of TiO2 are far from being completelyresolved, as was pointed out throughout this review. Some of the more important open questions are re-emphasized in the last few sections. There is ample need for addressing important, fundamental problems, as well as much excitement for signi®cant contributions that help to resolve these. Because TiO2 is used in so manydifferent ®elds it would be veryunfortunate if the rich and detailed knowledge that has been accumulated in recent years would go unnoticed by researchers in more applied areas. While some of these results reviewed here might turn out to be of mere fundamental interest and irrelevant for the particular environment, application or problem, some might help when trying to understand the behavior of this material. TiO2 will be important for years to come, and it is hoped that this review will help to link the more fundamental and more applied lines of research on this great material.

Acknowledgements

Foremost, I would like to thank Prof. Theodore E. Madeyfor stimulating myinitial interest in TiO 2, for his continued support throughout the years, and for his encouragement to write this review. I owe warm thanks to Dr. Michael A. Henderson for manyinsightful discussions and suggestions, and for giving me access to his comprehensive data base on TiO2 surfaces; to Dr. Matthias Batzill for critically reading this manuscript; to manycolleagues for providing me with unpublished results as well as reprints, preprints, and electronic versions of their ®gures; and to mystudents and post-docs for downloading and copying numerous articles and for help with ®gures and tables. The National Science Foundation, the Department of Energy, NASA, and the Louisiana Board of Regents are acknowledged for their continuing support of our work on TiO2 surfaces. I want to thank the Alexander von HumboldtÐFoundation for ®nancial support and Prof. H.J. Freund from the Fritz-Haber Institute, Germany, for his kind hospitality during the ®nal stages of writing this manuscript.

References

[1] V.E. Henrich, P.A. Cox, The Surface Science of Metal Oxides, Cambridge UniversityPress, Cambridge, 1994. [2] C. Noguera, Physics and Chemistry of Oxide Surfaces, Cambridge University Press, Cambridge, 1996. [3] H.-J. Freund, Angew. Chem. Int. Ed. Engl. 36 (1997) 452. [4] C.T. Campbell, Surf. Sci. Rep. 27 (1997) 1. [5] J.P. LaFemina, Crit. Rev. Surf. Chem. 3 (1994) 297. [6] M.A. Barteau, J. Vac. Sci. Technol. A 11 (1993) 2162. [7] D.A. Bonnell, Prog. Surf. Sci. 57 (1998) 187. [8] R.J. Lad, Surf. Rev. Lett. 2 (1995) 109. [9] R.J. Lad (Ed.), Surface Structure of Crystalline Ceramics, vol. 1, Elsevier, Amsterdam, 1996.

[10] R. Persaud, T.E. Madey, Growth, structure and reactivity of ultrathin metal films on TiO2 surfaces, in: D.A. King, D.P. Woodruff (Eds.), The Chemical Physics of Solid Surfaces, vol. 8, Elsevier, Amsterdam, 1997. [11] C.N. Satterfield, Heterogeneous Catalysis in Industrial Practice, 2nd ed., McGraw-Hill, New York, 1991. [12] J. Biener, J. Wang, R.J. Madix, Surf. Sci. 442 (1999) 47. U. Diebold / Surface Science Reports 48 (2003) 53±229 213

[13] Q. Guo, S. Lee, D.W. Goodman, Surf. Sci. 437 (1999) 38. [14] M. Sambi, G. Sangiovanni, G. Granozzi, F. Parmigiani, Phys. Rev. B 54 (1996) 13464. [15] Z. Zhang, V.E. Henrich, Surf. Sci. 277 (1992) 263. [16] G.L. Haller, D.E. Resasco, Adv. Catal. 36 (1989) 173. [17] J.M. Pan, T.E. Madey, Catal. Lett. 20 (1993) 269. [18] F. Pesty, H.P. SteinruÈck, T.E. Madey, Surf. Sci. 339 (1995) 83. [19] R.A. Bennett, P. Stone, M. Bowker, Catal. Lett. 59 (1999) 99. [20] O. Dulub, W. Hebenstreit, U. Diebold, Phys. Rev. Lett. 84 (2000) 3646. [21] M. Ando, T. Kobayashi, M. Haruta, Catal. Today 36 (1997) 135. [22] M. Valden, X. Lai, D.W. Goodman, Science 281 (1998) 1647. [23] V.A. Bondzie, S.C. Parker, C.T. Campbell, Catal. Lett. 63 (1999) 143. [24] L. Zhang, F. Cosandey, R. Persaud, T.E. Madey, Surf. Sci. 439 (1999) 73. [25] A. Fujishima, K. Honda, Nature 238 (1972) 37. [26] V.E. Henrich, G. Dresselhaus, H.J. Zeiger, Phys. Rev. Lett. 36 (1976) 1335. [27] W.J. Lo, Y.W. Chung, G.A. Somorjai, Surf. Sci. 71 (1978) 199. [28] H.O. Finklea, Semiconductor Electrodes, Elsevier, Amsterdam, 1988. [29] B. O'Regan, M. GraÈtzel, Nature 353 (1991) 737. [30] K. Rajeswar, J. Appl. Electrochem. 15 (1985) 1. [31] A. Mills, H.R. Davies, D. Worsley, Chem. Soc. Rev. 22 (1993) 417. [32] P.-C. Maness, S. Smolinski, W.A. Jacoby, Appl. Environ. Microbiol. 65 (1999) 4094. [33] Y. Paz, Z. Luo, L. Rabenberg, A. Heller, J. Mater. Res. 10 (1995) 2842. [34] I. Poulios, P. Spathis, P. Tsoumparis, J. Environ. Sci. Health 34 (1999) 1455. [35] R. Cai, K. Hashimoto, K. Itoh, Y. Kubota, A. Fujishima, Bull. Chem. Soc. Jpn. 64 (1991) 1268. [36] A. Fujishima, R. Cai, K. Hashimoto, H. Sakai, Y. Kubota, Trace Met. Environ. 3 (1993) 193. [37] H. Sakai, R. Baba, K. Hashimoto, Y. Kubota, A. Fujishima, Chem. Lett. (1995) 185. [38] O. Legrini, E. Oliveros, A.M. Braun, Chem. Rev. 93 (1993) 671. [39] D. Bahnemann, J. Cunningham, M.A. Fox, E. Pelizetti, P. Pichat, N. Serpone, Photocatalytic treatment of waters, in: G. Helz, R. Zepp, D. Crosby(Eds.), Aquatic and Surface Photochemistry,CRC Press, 1994, 261 pp. [40] D.Y. Goswami, Engineering of the solar photocatalytic detoxification and desinfection processes, in: K.W. Boer (Ed.), Advances in Solar Energy, vol. 10, American Solar Energy Society, Boulder, CO, 1995, 165 pp. [41] A. Heller, Acc. Chem. Res. 28 (1995) 503. [42] M. Hoffman, S. Martin, W. Choi, D. Bahnemann, Chem. Rev. 95 (1995) 69. [43] A.L. Linsebigler, G. Lu, J.T. Yates Jr., Chem. Rev. 95 (1995) 735. [44] G. Sheveglieri (Ed.), Gas Sensors, Kluwer Academic Publishers, Dordrecht, 1992. [45] P.K. Dutta, A. Ginwalla, B. Hogg, B.R. Patton, B. Chwieroth, Z. Liang, P. Gouma, M. Mills, S. Akbar, J. Phys. Chem. 103 (1999) 4412. [46] Y. Xu, K. Yao, X. Zhou, Q. Cao, Sens. Actuators B 13±14 (1993) 492. [47] U. Kirner, K.D. Schierbaum, W. GoÈpel, B. Leibold, N. Nicoloso, W. Weppner, D. Fischer, W.F. Chu, Sens. Actuators B 1 (1990) 103. [48] Kronos International, 1996. [49] F.A. Grant, Rev. Mod. Phys. 31 (1959) 646. [50] L.G. Phillips, D.M. Barbano, J. DairySci. 80 (1997) 2726. [51] J. Hewitt, Cosmet. Toiletries 114 (1999) 59. [52] H. Selhofer, Vacuum Thin Film (August 1999) 15. [53] E. Garfunkel, E. Gusev, A. Vul (Eds.), Fundamental Aspects of Ultrathin Dielectrics on Si-based Devices, NATO Science Series, Kluwer Academic Publishers, Dordrecht, 1998. [54] S.A. Campbell, H.-S. Kim, D.C. Gilmer, B. He, T. Ma, W.L. Gladfelter, IBM J. Res. Devlop. 43 (1999) 383. [55] Y. Matsumoto, T. Shono, T. Hasegawa, T. Fukumura, M. Kawasaki, P. Ahmet, T. Chikyow, S. Koshihara, H. Koinuma, Science 291 (2001) 854. [56] S.A. Chambers, S. Thevuthasan, R.F.C. Farrow, R.F. Marks, J.U. Thiele, L. Folks, M.G. Samant, A.J. Kellock, N. Ruzycki, D.L. Ederer, U. Diebold, Appl. Phys. Lett. 79 (2001) 3467. [57] P. Bonhote, E. Gogniat, M. GraÈtzel, P.V. Ashrit, Thin Solid Films 350 (1999) 269. 214 U. Diebold / Surface Science Reports 48 (2003) 53±229

[58] J.R. Sambrano, J. Andres, A. Beltran, F.R. Sensato, E.R. Leite, F.M.L.G. Stamato, E. Longo, Int. J. Quantum Chem. 65 (1997) 625. [59] B.D. Ratner, A.S. Hoffman, F.J. Schoen, J.E. Lemons (Eds.), Biomaterials ScienceÐAn Introduction to Materials in Medicine, Academic Press, San Diego, 1996. [60] C. Sittig, M. Textor, N.D. Spencer, M. Wieland, P.H. Vallotton, J. Mater. Sci. 10 (1999) 35. [61] J. Lausmaa, M. Ask, U. Rolander, B. Kasemo, Mater. Res. Soc. Symp. Proc. 110 (1988) 647. [62] P.W. Tasker, J. Phys. C 12 (1979) 4977. [63] U. Diebold, J.M. Pan, T.E. Madey, Surf. Sci. 333 (1995) 845. [64] N.M. Harrison, X.G. Wang, J. Muscat, M. Scheffler, FaradayDiscuss., Chem. Soc. 114 (1999) 305. [65] G.V. Samsonov, The Oxide Handbook, IFI/Plenum Press, New York, 1982. [66] L.S. Dubrovinsky, N.A. Dubrovinskaia, V. Swamy, J. Muscat, N.M. Harrison, R. Ahuja, B. Holm, B. Johansson, Nature 410 (2001) 653. [67] U. Diebold, Specimen treatment: preparation of metal compound materials (mainlyoxides), in: C.A. Czanderna, C.J. Powell, T.E. Madey(Eds.), Specimen Handling, Treatments, Beam Effects and Depth Profiling, vol. 4, 1999. [68] M. Ramamoorthy, D. Vanderbilt, Phys. Rev. B 49 (1994) 16721. [69] A. Zangwill, Physics at Surfaces, Cambridge, 1988. [70] A. Vittadini, A. Selloni, F.P. Rotzinger, M. GraÈtzel, Phys. Rev. Lett. 81 (1998) 2954. [71] U. Diebold, M. Li, O. Dulub, E.L.D. Hebenstreit, W. Hebenstreit, Surf. Rev. Lett. 5±6 (2000) 613. [72] K.D. Schierbaum, S. Fischer, M.C. Torquemada, J.L. de Segovia, E. RomaÂn, J.A. Martin-Gato, Surf. Sci. 345 (1996) 261. [73] M.A. Henderson, Surf. Sci. 343 (1995) L1156. [74] M.A. Henderson, Surf. Sci. 419 (1999) 174. [75] M. Li, W. Hebenstreit, L. Gross, U. Diebold, M.A. Henderson, D.R. Jennison, P.A. Schultz, M.P. Sears, Surf. Sci. 437 (1999) 173. [76] R.A. Bennett, P. Stone, N.J. Price, M. Bowker, Phys. Rev. Lett. 82 (1999) 3831. [77] E.L.D. Hebenstreit, W. Hebenstreit, U. Diebold, Surf. Sci. 461 (2000) 87. [78] E. Yagi, R. Hasiguti, M. Aono, Phys. Rev. B 54 (1996) 7945. [79] R.R. Hasiguti, E. Yagi (1994). [80] P. Kofstad, J. Less-Common Met. 13 (1967) 635. [81] D.J. Smith, L.A. Bursill, M.G. Blanchin, Philisophical Mag. A 50 (1984) 473. [82] L.A. Bursill, D.J. Smith, Nature 309 (1984) 319. [83] L.A. Bursill, M.G. Blanchin, D.J. Smith, Proc. Roy. Soc. London Ser. A 391 (1984) 351. [84] G.S. Rohrer, V.E. Henrich, D.A. Bonnell, Science 250 (1990) 1239. [85] H. NoÈrenberg, R.E. Tanner, K.D. Schierbaum, S. Fischer, G.A.D. Briggs, Surf. Sci. 396 (1998) 52. [86] H. NoÈrenberg, G.A.D. Briggs, Surf. Sci. 404 (1998) 738. [87] R.A. Bennett, S. Poulston, P. Stone, M. Bowker, Phys. Rev. B 59 (1999) 10341. [88] P.W. Murray, N.G. Condon, G. Thornton, Phys. Rev. B 51 (1995) 10989. [89] J. Sasaki, N.L. Peterson, K. Hoshino, J. Phys. Chem. Solids 46 (1985) 1267. [90] H.B. Huntington, G.A. Sullivan, Phys. Rev. Lett. 14 (1965) 177. [91] S.D. Elliott, S.P. Bates, Phys. Chem. Chem. Phys. 3 (2001) 1954. [92] R. Wang, K. Hashimoto, A. Fujishima, M. Chikuni, E. Kojima, A. Kitamura, M. Shimohigoshi, T. Watanabe, Nature 388 (1997) 431. [93] Y. Gao, S.A. Chambers, Mater. Res. Soc. Symp. Proc. 401 (1996) 85. [94] G. Charlton, P.B. Hoowes, C.L. Nicklin, P. Steadman, J.S.G. Taylor, C.A. Muryn, S.P. Harte, J. Mercer, R. McGrath, D. Norman, T.S. Turner, G. Thornton, Phys. Rev. Lett. 78 (1997) 495. [95] B. Hird, R.A. Armstrong, Surf. Sci. 420 (1999) L131. [96] B. Hird, R.A. Armstrong, Surf. Sci. 385 (1997) L1023. [97] E. Asari, T. Suzuki, R. Souda, Phys. Rev. B 61 (2000) 5679. [98] A. Verdini, M. Sambi, F. Bruno, D. Cvetko, M.D. Negra, R. Gotter, L. Floreano, A. Morgante, G.A. Rizzi, G. Granozzi, Surf. Rev. Lett. 6 (1999) 2101. [99] P.J.D. Lindan, J. Muscat, S. Bates, N.M. Harrison, M. Gillan, FaradayDiscuss., Chem. Soc. 106 (1997) 135. [100] D. Vogtenhuber, R. Podloucky, A. Neckel, S.G. Steinemann, A.J. Freeman, Phys. Rev. B 49 (1994) 2099. U. Diebold / Surface Science Reports 48 (2003) 53±229 215

[101] P. Reinhardt, B.A. Heû, Phys. Rev. B 50 (1994) 12015. [102] S.P. Bates, G. Kresse, M.J. Gillan, Surf. Sci. 409 (1998) 336. [103] D.R. Hamann, Phys. Rev. B 56 (1997) 14979. [104] K.M. Glassford, J.R. Chelikowsky, Phys. Rev. B 46 (1992) 1284. [105] W. Langel, Surf. Sci. 496 (2002) 141. [106] D. Vogtenhuber, R. Poducky, J. Redinger, E.L.D. Hebenstreit, W. Hebenstreit, U. Diebold, Phys. Rev. B 65 (2002) 125411/1. [107] G. Charlton, P.B. Howes, C.A. Muryn, H. Raza, N. Jones, J.S.G. Taylor, C. Norris, R. McGrath, D. Norman, T.S. Turner, G. Thornton, Phys. Rev. B 61 (2000) 16117. [108] H. Onishi, Y. Iwasawa, Chem. Phys. Lett. 226 (1994) 111. [109] U. Diebold, W. Hebenstreit, G. Leonardelli, M. Schmid, P. Varga, Phys. Rev. Lett. 81 (1998) 405. [110] U. Diebold, J.F. Anderson, K.O. Ng, D. Vanderbilt, Phys. Rev. Lett. 77 (1996) 1322. [111] P.J. Mùller, M.-C. Wu, Surf. Sci. 224 (1989) 265. [112] O. GuÈlseren, R. James, D.W. Bullett, Surf. Sci. 377±379 (1997) 150. [113] R.E. Tanner, M.R. Castell, G.A.D. Briggs, Surf. Sci. 412 (1998) 672. [114] A. Szabo, T. Engel, Surf. Sci. 329 (1995) 241. [115] Q. Guo, I. Cocks, E.M. Williams, J. Phys. D 31 (1998) 2231. [116] U. Diebold, J. Lehman, T. Mahmoud, M. Kuhn, G. Leonardelli, W. Hebenstreit, M. Schmid, P. Varga, Surf. Sci. 411 (1998) 137. [117] S. Fischer, A.W. Munz, K.D. Schierbaum, W. GoÈpel, Surf. Sci. 337 (1995) 17. [118] K.-i. Fukui, H. Onishi, Y. Iwasawa, Phys. Rev. Lett. 79 (1997) 4202. [119] T.R. Albrecht, P. GruÈtter, D. Horne, D. Rugar, J. Appl. Phys. 69 (1991) 668. [120] M. Reichling, C. Barth, Phys. Rev. Lett. 83 (1999) 768. [121] B. Grossmann, P. Piercy, Phys. Rev. Lett. 74 (1995) 4487. [122] H. Onishi, K.-i. Fukui, Y. Iwasawa, Bull. Chem. Soc. Jpn. 68 (1995) 2447. [123] H. Onishi, Y. Iwasawa, Surf. Sci. 313 (1994) L783. [124] S. Suzuki, Y. Yamaguchi, H. Onishi, K.-i. Fukui, T. Sasaki, Y. Iwasawa, Catal. Lett. 50 (1998) 117. [125] S. Gan, Y. Liang, D.R. Baer, Phys. Rev. B 63 (2001) 121401/1. [126] Y. Iwasawa, H. Onishi, K.-i. Fukui, S. Suzuki, T. Sasaki, FaradayDiscuss., Chem. Soc. 114 (1999) 259. [127] J.M. Pan, B.L. Maschhoff, U. Diebold, T.E. Madey, J. Vac. Sci. Technol. A 10 (1992) 2470. [128] M.A. Henderson, Surf. Sci. 355 (1996) 151. [129] C.A. Jenkins, D.M. Murphy, J. Phys. Chem. B 103 (1999) 1019. [130] L.Q. Wang, D.R. Baer, M.H. Engelhard, Surf. Sci. 320 (1994) 295. [131] W. GoÈpel, G. Rocker, R. Feierabend, Phys. Rev. B 28 (1983) 3427. [132] D. Vogtenhuber, Private communication. [133] S. Suzuki, K.-i. Fukui, H. Onishi, Y. Iwasawa, Phys. Rev. Lett. 84 (2000) 2156. [134] R. Schaub, P. Thostrup, N. Lopez, E. Lñgsgaard, I. Stensgaard, J.K. Nùrskov, F. Besenbacher, Phys. Rev. Lett. 87 (2001) 266104/1. [135] M.L. Knotek, P.J. Feibelman, Surf. Sci. 90 (1979) 78. [136] U. Diebold, T.E. Madey, Phys. Rev. Lett. 72 (1994) 1116. [137] E. Bertel, R. Stockbauer, T.E. Madey, Surf. Sci. 141 (1984) 355. [138] L.Q. Wang, D.R. Baer, M.H. Engelhard, A.N. Shultz, Surf. Sci. 344 (1995) 237. [139] S.A. Joyce, Private communication. [140] A. BerkoÂ, E. Krivan, J. Vac. Sci. Technol. B 15 (1997) 25. [141] M.R. McCartney, D.J. Smith, Surf. Sci. 250 (1991) 169. [142] A.N. Shultz, W. Jang, W.M.I. Hetherington, D.R. Baer, L.Q. Wang, M.H. Engelhard, Surf. Sci. 339 (1995) 114. [143] T.E. Madey, B.L. Maschhoff, Unpublished results. [144] M. Li, W. Hebenstreit, U. Diebold, Phys. Rev. B 61 (2000) 4926. [145] L.P. Zhang, M. Li, U. Diebold, Surf. Sci. 412 (1998) 242. [146] H. NoÈrenberg, J.H. Harding, Appl. Surf. Sci. 142 (1999) 174. [147] H. NoÈrenberg, J.H. Harding, Phys. Rev. B 59 (1999) 9842. [148] H. NoÈrenberg, J.H. Harding, Surf. Sci. 473 (2001) 151. 216 U. Diebold / Surface Science Reports 48 (2003) 53±229

[149] J.F. Anderson, M. Kuhn, U. Diebold, K. Shaw, P. Stoyanov, D. Lind, Phys. Rev. B 56 (1997) 9902. [150] S. Gan, Y. Liang, D.R. Baer, Surf. Sci. 459 (2000) L498. [151] M.A. Henderson, S. Otero-Tapia, M.E. Castro, Surf. Sci. 412 (1998) 252. [152] M.A. Henderson, Private communication. [153] M. Batzill, B. Katsiev, D.J. Gaspar, U. Diebold, Phys. Rev. B 66 (2002). [154] G.S. Rohrer, V.E. Henrich, D.A. Bonnell, Surf. Sci. 278 (1992) 146. [155] H. NoÈrenberg, G.A.D. Briggs, Surf. Sci. 404 (1998) 738. [156] M. Li, W. Hebenstreit, U. Diebold, A.M. Tyryshkin, M.K. Bowman, G.G. Dunham, M.A. Henderson, J. Phys. Chem. B 104 (2000) 4944. [157] R.A. Bennett, Phys. Chem. Commun. 3 (2000) (web-based journal). [158] M. Sander, T. Engel, Surf. Sci. 302 (1994) L263. [159] M. Wagner, O. Kienzle, D.A. Bonnell, M. RuÈhle, J. Vac. Sci. Technol. A 16 (1998) 1078. [160] P.W. Murray, N.G. Condon, G. Thornton, Surf. Sci. 323 (1995) L281. [161] A. BerkoÂ, F. Solymosi, Langmuir 12 (1996) 1257. [162] C. Xu, X. Lai, G.W. Zajac, D.W. Goodman, Phys. Rev. B 56 (1997) 13464. [163] C.L. Pang, S.A. Haycock, H. Raza, P.W. Murray, G. Thornton, O. GuÈlseren, R. James, D.W. Bullett, Phys. Rev. B 58 (1998) 1586. [164] E. Asari, R. Souda, Phys. Rev. B 60 (1999) 10719. [165] K.O. Ng, D. Vanderbilt, Phys. Rev. B 56 (1997) 10544. [166] Q. Guo, I. Cocks, E.M. Williams, Phys. Rev. Lett. 77 (1996) 3851. [167] R.E. Tanner, M.R. Castell, G.A.D. Briggs, Surf. Sci. 437 (1999) 263. [168] C.L. Pang, S.A. Haycock, H. Raza, G. Thornton, O. GuÈlseren, R. James, D.W. Bullet, Surf. Sci. 437 (1999) 261. [169] H. Onishi, Y. Iwasawa, Phys. Rev. Lett. 76 (1996) 791. [170] M. Li, W. Hebenstreit, U. Diebold, Surf. Sci. 414 (1998) L951. [171] M. Li, W. Hebenstreit, U. Diebold, M.A. Henderson, D.R. Jennison, FaradayDiscuss., Chem. Soc. 114 (1999) 245. [172] R.A. Bennett, P. Stone, M. Bowker, FaradayDiscuss., Chem. Soc. 114 (1999) 267. [173] P. Stone, R.A. Bennett, M. Bowker, New J. Phys. 1 (1999) 8. [174] M.A. Henderson, Langmuir 12 (1996) 5093. [175] M.B. Hugenschmidt, L. Gamble, C.T. Campbell, Surf. Sci. 302 (1994) 329. [176] H. Raza, C.L. Pang, S.A. Haycock, G. Thornton, Phys. Rev. Lett. 82 (1999) 5265. [177] H. Raza, C.L. Pang, S.A. Haycock, G. Thornton, Appl. Surf. Sci. 140 (1999) 271. [178] J. Muscat, N.M. Harrison, G. Thornton, Phys. Rev. B 59 (1999) 2320. [179] P.J.D. Lindan, N.M. Harrison, J.M. Holender, M.J. Gillan, M.C. Payne, Surf. Sci. 364 (1996) 431. [180] P.J. Hardman, P.L. Wincott, Phys. Rev. B 60 (1999) 11700. [181] Y.W. Chung, W.J. Lo, G.A. Somorjai, Surf. Sci. 64 (1977) 588. [182] P.J. Hardman, N.S. Prakash, C.A. Muryn, G.N. Raikar, A.G. Thomas, R.J. Blake, Phys. Rev. B 47 (1993) 16056. [183] C.A. Muryn, G. Tirvengadum, J.J. Crouch, D.R. Warburton, G.N. Raiker, G. Thornton, D.S.-L. Law, J. Phys. C. 1 suppl. B (1991) 265. [184] M.C. Wu, P.J. Mùller, Surf. Sci. 224 (1989) 250. [185] G.W. Clark, L.L. Kesmodel, Ultramicroscopy41 (1992) 77. [186] P. Zschack, J.B. Cohen, Y.W. Chung, Surf. Sci. 262 (1992) 395. [187] P.W. Murray, F.M. Leibsle, H.J. Fisher, C.F.J. Flipse, C.A. Muryn, G. Thornton, Phys. Rev. Lett. 46 (1992) 12877. [188] P.W. Murray, F.M. Leibsle, C.A. Muryn, H.J. Fisher, C.F.J. Flipse, G. Thornton, Phys. Rev. Lett. 72 (1994) 689. [189] P.W. Murray, F.M. Leibsle, C.A. Muryn, C.F.J. Flipse, G. Thornton, Surf. Sci. 321 (1994) 217. [190] Y. Gao, Y. Liang, S.A. Chambers, Surf. Sci. 365 (1996) 638. [191] P.M. Oliver, S.C. Parker, J. Purton, D.W. Bullett, Surf. Sci., Part B 307±309 (1994) 1200. [192] H. Zajonz, H.L. Meyerheim, T. Gloege, W. Moritz, D. Wolf, Surf. Sci. 398 (1998) 369. [193] E. Landree, L.D. Marks, P. Zschack, C.J. Gilmore, Surf. Sci. 408 (1998) 300. [194] L.D. Marks, R. Plass, D.L. Dorset, Surf. Rev. Lett. 4 (1997) 1. [195] Q. Guo, I. Cocks, E.M. Williams, Surf. Sci. 366 (1996) 99. [196] M.A. Henderson, J. Phys. Chem. 99 (1995) 15253. [197] M.A. Henderson, Surf. Sci. 319 (1994) 315. U. Diebold / Surface Science Reports 48 (2003) 53±229 217

[198] R.H. Tait, R.V. Kasowski, Phys. Rev. B 20 (1979) 5178. [199] L.E. Firment, Surf. Sci. 116 (1982) 205. [200] G.E. Poirier, B.K. Hance, M. White, J. Vac. Sci. Technol. B 10 (1992) 6. [201] B.A. Watson, M.A. Barteau, Chem. Mater. 6 (1994) 771. [202] K.-i. Fukui, R. Tero, Y. Iwasawa, Jpn. J. Appl. Phys., Part 1 40 (2001) 4331. [203] H. NoÈrenberg, F. Dinelli, G.A.D. Briggs, Surf. Sci. 446 (2000) L83. [204] H. NoÈrenberg, F. Dinelli, G.A.D. Briggs, Surf. Sci. 436 (1999) L635. [205] H. Onishi, T. Aruga, C. Egawa, Y. Iwasawa, Surf. Sci. 199 (1988) 54. [206] A. Howard, C.E.J. Mittchell, D. Morris, R.G. Egdell, S.C. Parker, Surf. Sci. 448 (2000) 131. [207] H. Uetsuka, A. Sasahara, H. Onishi, Jpn. J. Appl. Phys., Part 1 39 (2000) 3769. [208] T. Ohno, K. Sarukawa, M. Matsumura, J. Phys. Chem. B 105 (2001) 2417. [209] L. Kavan, M. GraÈtzel, S.E. Gilbert, C. Klemenz, H.J. Scheel, J. Am. Chem. Soc. 118 (1996) 6716. [210] J.M.G. Amores, V.S. Escribano, G. Busca, J. Mater. Chem. 5 (1995) 1245. [211] C. Byun, J.W. Jang, I.T. Kim, K.S. Hong, B.-W. Lee, Mater. Res. Bull. 32 (1997) 431. [212] P. Arnal, R.J.P. Corriu, D. Leclercq, P.H. Mutin, A. Vioux, J. Mater. Chem. 6 (1996) 1925. [213] T. Oyama, Y. Iimura, K. Takeuchi, T. Ishii, J. Mater. Sci. Lett. 15 (1996) 594. [214] A. Fahmi, C. Minot, Surf. Sci. 304 (1994) 343. [215] T. Bredow, K. Jug, Surf. Sci. 327 (1995) 398. [216] M. Lazzeri, A. Vittadini, A. Selloni, Phys. Rev. B 63 (2001) 155409/1. [217] M. Lazzeri, A. Vittadini, A. Selloni, Phys. Rev. B 65 (2002) 119901/1. [218] G.S. Herman, Y. Gao, T.T. Tran, J. Osterwalder, Surf. Sci. 447 (1999) 201. [219] G.S. Herman, M.R. Sievers, Y. Gao, Phys. Rev. Lett. 84 (2000) 3354. [220] W. Hebenstreit, N. Ruzycki, G.S. Herman, Y. Gao, U. Diebold, Phys. Rev. B 64 (2000) R16344. [221] G.S. Herman, Y. Gao, Thin Solid Films 397 (2001) 157. [222] R. Hengerer, B. Bolliger, M. Erbudak, M. GraÈtzel, Surf. Sci. 460 (2000) 162. [223] J. Woning, R.A. van Santen, Chem. Phys. Lett. 101 (1983) 541. [224] S. Chen, M.G. Mason, H.J. Gysling, G.R. Paz-Pajult, T.N. Blanton, K.M. Chen, C.P. Fictorie, W.L. Gladfelter, A. Franciosi, P.I. Cohen, J.F. Evans, J. Vac. Sci. Technol. A 11 (1993) 2419. [225] D.S. Lind, S.D. Berry, G. Chern, H. Mathias, L.R. Testardi, Phys. Rev. B 45 (1992) 1838. [226] W. Sugimura, A. Yamazaki, H. Shigetani, J. Tanaka, T. Mitsuhashi, Jpn. J. Appl. Phys. 36 (1997) 7358. [227] R. Hengerer, L. Kavan, M. GraÈtzel, J. Electrochem. Soc. 147 (2000) 1467. [228] Y. Liang, S. Gan, S.A. Chambers, E.I. Altman, Phys. Rev. B 63 (2001) 235402/1. [229] R.E. Tanner, A. Sasahara, Y. Liang, E.I. Altman, H. Onishi, J. Phys. Chem. B 106 (2002) 8211. [230] M. Lazzeri, A. Selloni, Phys. Rev. Lett. 87 (2001) 266105/1. [231] N. Ruzycki, G. Herman, L.A. Boatner, U. Diebold, Surf. Sci., submitted. [232] P.A. Cox, Transition Metal OxidesÐAn Introduction to Their Electronic Structure, Properties, Clarendon Press, Oxford, 1992. [233] P.J.D. Lindan, N.M. Harrison, M.J. Gillan, J.A. White, Phys. Rev. B 55 (1997) 15919. [234] D. Vogtenhuber, R. Podloucky, J. Redinger, E.L. Bullock, L. Patthey, S.G. Steinemann, Proc. Electrochem. Soc. 15 (1994) 96. [235] J. Goniakowski, C. Noguera, Surf. Sci. 319 (1994) 68. [236] J. Purton, D.W. Bullett, P.M. Oliver, S.C. Parker, Surf. Sci. 336 (1995) 166. [237] N. Yu, J.W. Halley, Phys. Rev. B 51 (1995) 4768. [238] J. Goniakowski, C. Noguera, Surf. Sci. 323 (1995) 129. [239] K.D. Schierbaum, W.X. Xu, Int. J. Quantum Chem. 57 (1996) 1121. [240] S. Kimura, M. Tsukada, Appl. Surf. Sci. 132 (1998) 587. [241] K.F. Ferris, L.Q. Wang, J. Vac. Sci. Technol. A 16 (1998) 956. [242] M. Casarin, C. Maccato, A. Vittadini, J. Phys. Chem. B 102 (1998) 10745. [243] P.K. Schelling, N. Yu, J.W. Halley, Phys. Rev. B 58 (1998) 1279. [244] F. Rittner, R. Fink, B. Boddenberg, V. Staemmler, Phys. Rev. B 57 (1998) 4160. [245] A.T. Paxton, L. ThieÃn-Nga, Phys. Rev. B 57 (1998) 1579. [246] S. Matsushima, K. Kobayashi, M. Kohyama, Kitakyushu Kogyo Koto Senmon Gakko Kenkyu Hokoku 32 (1999) 155. 218 U. Diebold / Surface Science Reports 48 (2003) 53±229

[247] T. Albaret, F. Finocchi, C. Noguera, FaradayDiscuss., Chem. Soc. 114 (1999) 285. [248] M. Ramamoorthy, R.D. King-Smith, D. Vanderbilt, Phys. Rev. B 49 (1994) 7709. [249] S.P. Bates, G. Kresse, M.J. Gillan, Surf. Sci. 385 (1997) 386. [250] P.J.D. Lindan, N.M. Harrison, Surf. Sci. 479 (2001) L375. [251] M. Tsukada, C. Satoko, H. Adachi, J. Phys. Soc. Jpn. 44 (1978) 1043. [252] R.V. Kasowski, R.H. Tait, Phys. Rev. B 20 (1979) 5168. [253] M. Tsukada, C. Satoko, H. Adachi, J. Phys. Soc. Jpn. 47 (1979) 1610. [254] S. Munnix, M. Schmeits, Surf. Sci. 126 (1983) 20. [255] S. Munnix, M. Schmeits, Phys. Rev. B 28 (1983) 7342. [256] S. Munnix, M. Schmeits, Phys. Rev. B 30 (1984) 2202. [257] S. Munnix, M. Schmeits, Phys. Rev. B 31 (1985) 3369. [258] K.M. Glassford, N. Troullier, J.L. Martins, J.R. Chelikowsky, Sol. State Commun. 76 (1990) 635. [259] R. Heise, R. Courths, S. Witzel, Solid State Commun. 84 (1992) 599. [260] R. Heise, R. Courths, Springer Ser. Surf. Sci. 33 (1993) 91. [261] Z. Zhang, S.P. Jeng, V.E. Henrich, Phys. Rev. B 43 (1991) 12004. [262] U. Diebold, H.S. Tao, N.D. Shinn, T.E. Madey, Phys. Rev. B 50 (1994) 14474. [263] R. Heise, R. Courths, Surf. Sci. 288 (1993) 658. [264] A.E. Taverner, P.C. Hollamby, P.S. Aldridge, R.G. Egdell, W.C. Mackrodt, Surf. Sci. 288 (1993) 653. [265] A.K. See, R.A. Bartynski, J. Vac. Sci. Technol. A 10 (1992) 2591. [266] A.K. See, M. Thayer, R.A. Bartynski, Phys. Rev. B 47 (1993) 13722. [267] L. Soriano, M. Abbate, J. Vogel, J.C. Fuggle, A. Fernandez, A.R. Gonzalez-Elipe, M. Sacchi, J.M. Sanz, Surf. Sci. 290 (1993) 427. [268] R. Brydson, B.G. Williams, W. Engel, H. Sauer, E. Zeitler, J.M. Thomas, Solid State Commun. 64 (1987) 609. [269] F.M.F. de Groot, J. Electr. Spectrosc. Rel. Phen. 67 (1993) 529. [270] F.M.F. de Groot, J. Electr. Spectrosc. Rel. Phen. 62 (1993) 111. [271] U. Diebold, N.D. Shinn, Surf. Sci. 343 (1995) 53. [272] J. Biener, M. BaÈumer, R.J. Madix, Surf. Sci. 432 (1999) 178. [273] P.J. Hardman, G.N. Raikar, C.A. Muryn, G. van der Laan, P.L. Wincott, G. Thornton, D.W. Bullett, P.A.D.M.A. Dale, Phys. Rev. B 49 (1994) 7170. [274] Y.Aiura, Y.Nishihara, Y.Haruyama, T. Komeda, S. Kodaira, Y.Sakisaka, T. Maruyama, H. Kato, Physica B 196 (1994) 1215. [275] M.A. Henderson, W.S. Epling, C.L. Perkins, C.H.F. Peden, U. Diebold, J. Phys. Chem. B 103 (1999) 5328. [276] W. GoÈpel, J.A. Anderson, D. Frankel, M. Jaehnig, K. Phillips, J.A. SchaÈfer, G. Rocker, Surf. Sci. 139 (1984) 333. [277] M.A. Henderson, Surf. Sci. 400 (1998) 203. [278] J. Muscat, N.M. Harrison, G. Thornton, Phys. Rev. B 59 (1999) 15457. [279] U. Diebold, T.E. Madey, Surf. Sci. Spectra 4 (1998) 227. [280] J.T. Mayer, U. Diebold, T.E. Madey, E. Garfunkel, J. Electr. Spectrosc. Rel. Phen. 73 (1995) 1. [281] L.S. Dake, R.J. Lad, Surf. Sci. Spectra 4 (1998) 232. [282] A.E. Bocquet, T. Mizokawa, K. Morikawa, A. Fujimori, S.R. Barman, K. Maiti, D.D. Sarma, Y. Tokura, M. Onoda, Phys. Rev. B 53 (1996) 1161. [283] G. Rocker, J.A. SchaÈfer,W.GoÈpel, Phys. Rev. B 30 (1984) 3704. [284] L.L. Kesmodel, J.A. Gates, Y.W. Chung, Phys. Rev. B 23 (1981) 489. [285] P.A. Cox, R.G. Egdell, S. Eriksen, W.R. Flavell, J. Electr. Spectrosc. Rel. Phen. 39 (1986) 117. [286] S. Eriksen, R.G. Egdell, Surf. Sci. 180 (1987) 263. [287] M.-C. Wu, C.A. Estrada, D.W. Goodman, Phys. Rev. Lett. 67 (1991) 2910. [288] P.A. Cox, A.A. Williams, Surf. Sci. 152±153 (1985) 791. [289] S.A. Chambers, Surf. Sci. Rep. 39 (2000) 105. [290] L. ThieÃn-Nga, A.T. Paxton, Phys. Rev. B 58 (1998) 13233. [291] S.H. Overbury, P.A. Bertrand, G.A. Somorjai, Chem. Rev. 75 (1975) 547. [292] J.M. Pan, T.E. Madey, J. Vac. Sci. Technol. A 11 (1993) 1667. [293] C.R. Henry, Surf. Sci. Rep. 31 (1998) 231. [294] F. Cosandey, T.E. Madey, Surf. Rev. Lett. 8 (2001) 73. [295] D.A. Chen, M.C. Bartelt, K.F. McCarty, Surf. Sci. 450 (2000) 78. U. Diebold / Surface Science Reports 48 (2003) 53±229 219

[296] O. Dulub, L.A. Boatner, U. Diebold, Surf. Sci. 504 (2002) 271. [297] M.J.J. Jak, C. Konstapel, A. van Kreuningen, J. Chrost, J. Verhoeven, J.W.M. Frenken, Surf. Sci. 474 (2001) 28. [298] S.J. Tauster, Strong-metal support interactions. Facts and uncertainties, in: Proceedings of the Strong Metal±Support Interactions, Symposium at the 189th Meeting of the American Chemical Society, Miami Beach, FL, 1986. [299] A.K. Datye, D.S. Kalakkad, M.H. Yao, J. Catal. 155 (1995) 148.

[300] M.-H. Yao, HREM studyof strong metal±support interaction in Pt/TiO 2, in: Proceedings of the 51st Annual Meeting of the MicroscopySocietyof America, 1993, San Francisco. [301] P.L.J. Gunter, J.W. Niemandsverdriet, G.A. Somorjai, Catal. Rev., Sci. Eng. 39 (1997) 77. [302] D.R. Jennison, O. Dulub, W. Hebenstreit, U. Diebold, Surf. Sci. 492 (2001) L677. [303] M. Bowker, J. Phys. Chem. 106 (2002) 4688. [304] H. Onishi, T. Aruga, C. Egawa, Y. Iwasawa, Shokubai 29 (1987) 518. [305] H. Onishi, T. Aruga, C. Egawa, Y. Iwasawa, J. Chem. Soc., FaradayTrans. 85 (1989) 2597. [306] F. Cosandey, R. Persaud, L. Zhang, T.E. Madey, Mater. Res. Soc. Symp. Proc. 440 (1997) 383. [307] A. Kolmakov, D.W. Goodman, Catal. Lett. 70 (2000) 93. [308] X. Lai, T.P.S. Clair, D.W. Goodman, FaradayDiscuss., Chem. Soc. 114 (1999) 279. [309] A. Kolmakov, D.W. Goodman, Surf. Sci. 490 (2001) L597. [310] A. Kolmakov, D.W. Goodman, Catal. Lett. 70 (2000) 93. [311] M. GraÈtzel, MRS Bull. (October 1993) 61. [312] M.K. Jain, M.C. Bhatnagar, G.L. Sharma, Sens. Actuators 55 (1999) 180. [313] A.M. Efstathiou, D. Boudovas, N. Vamvouka, X.E. Verykios, J. Catal. 140 (1993) 1. [314] V.N. Ageev, S.M. Solov'ev, Phys. Solid State 42 (2000) 2159. [315] V.N. Ageev, S.M. Solov'ev, Surf. Sci. 480 (2001) 1. [316] A. Stashans, S. Lunell, R. Bergstroem, A. Hagfeldt, S.E. Lindquist, Phys. Rev. B 53 (1996) 159. [317] R. van de Krol, A. Goossens, E.A. Meulenkamp, J. Electrochem. Soc. 146 (1999) 3150. [318] R. van de Krol, A. Goossens, J. Schoonman, J. Phys. Chem. 103 (1999) 7151. [319] M.A. San Miguel, C.J. Calzado, J.F. Sanz, J. Phys. Chem. B 105 (2001) 1794. [320] Y. Iwasawa, Acid Base Catal., Proc. Int. Symp. (1989) 267. [321] R. Souda, W. Hayami, T. Aizawa, Y. Ishizawa, Phys. Rev. B 48 (1993) 17255. [322] R. Souda, W. Hayami, T. Aizawa, Y. Ishizawa, Surf. Sci. 285 (1993) 265. [323] H. Onishi, Y. Iwasawa, Catal. Lett. 38 (1996) 89. [324] J. Nerlov, Q. Ge, P.J. Mùller, Surf. Sci. 348 (1996) 28. [325] P.J. Mùller, NATO ASI Ser., Ser. E 331 (1997) 267. [326] J. Nerlov, S.V. Christensen, S. Weichel, E.H. Pedersen, P.J. Mùller, Surf. Sci. 371 (1997) 321. [327] B. Hird, R.A. Armstrong, Surf. Sci. 431 (1999) L570. [328] A. Gutierrez-Sosa, J.F. Walsh, R. Lindsay, P.L. Wincott, G. Thornton, Surf. Sci. 435 (1999) 538. [329] M.A. San Miguel, C.J. Calzado, J.F. Sanz, Surf. Sci. 409 (1998) 92. [330] M.A. San Miguel, C.J. Calzado, J.F. Sanz, Int. J. Quantum Chem. 70 (1998) 351. [331] J.F. Sanz, C.M. Zicovich-Wilson, Chem. Phys. Lett. 303 (1999) 111. [332] T. Albaret, F. Finocchi, C. Noguera, A. De Vita, Phys. Rev. B 65 (2002) 035402/1. [333] R. Heise, R. Courths, Surf. Rev. Lett. 2 (1995) 147. [334] R. Heise, R. Courths, Surf. Sci. 333 (1995) 1460. [335] R.J. Lad, L.S. Dake, Mater. Res. Soc. Symp. Proc. 238 (1992) 823. [336] T. Bredow, Int. J. Quantum Chem. 75 (1999) 127. [337] D. Purdie, B. Reihl, G. Thornton, J. de phys. IV 4 (1994) 163. [338] G. Thornton, Vacuum 43 (1992) 1133. [339] K. Prabhakaran, D. Purdie, R. Casanova, C.A. Muryn, P.J. Hardman, P.L. Wincott, G. Thornton, Phys. Rev. B 45 (1992) 6969. [340] B.E. Hayden, G.P. Nicholson, Surf. Sci. 274 (1992) 277. [341] A.G. Thomas, P.J. Hardman, C.A. Muryn, H.S. Dhariwal, A.F. Prim, G. Thornton, E. RomaÂn, J. de Segovia, J. Chem. Soc., FaradayTrans. 91 (1995) 3569. [342] M. Brause, V. Kempter, Surf. Sci. 476 (2001) 78. [343] A.W. Grant, C.T. Campbell, Phys. Rev. B 55 (1997) 1844. 220 U. Diebold / Surface Science Reports 48 (2003) 53±229

[344] M. Brause, S. Skordas, V. Kempter, Surf. Sci. 445 (2000) 224. [345] L.S. Dake, R.J. Lad, Surf. Sci. 289 (1993) 297. [346] X. Lai, T.P.S. Clair, M. Valden, D.W. Goodman, Prog. Surf. Sci. 59 (1998) 25. [347] X. Lai, C. Xu, D.W. Goodman, J. Vac. Sci. Technol. A 16 (1998) 2562. [348] D.L. Carroll, Y. Liang, D.A. Bonnell, J. Vac. Sci. Technol. A 12 (1994) 2298. [349] L.S. Dake, R.J. Lad, J. Vac. Sci. Technol. A 13 (1995) 122. [350] G. Rocker, W. GoÈpel, Surf. Sci. 181 (1987) 530. [351] U. Diebold, T.E. Madey, Unpublished results. [352] N.J. Price, J.B. Reitz, R.J. Madix, E.I. Solomon, J. Electr. Spectrosc. Rel. Phen. 99 (1999) 257. [353] M. Sambi, E. Pin, G. Saniovanni, L. Zaratin, G. Granozzi, F. Parmigiani, Surf. Sci. 349 (1996) L169. [354] J. Biener, M. BaÈumer, R. Madix, Surf. Sci. 450 (2000) 12. [355] J.M. Pan, B.L. Maschhoff, U. Diebold, T.E. Madey, in: Proceedings of the Fourth International Conference on Struct. Surf. IV, 1994, p. 613. [356] R.J. Madix, J. Biener, M. BaÈumer, A. Dinger, FaradayDiscuss., Chem. Soc. 114 (1999) 67. [357] J. Wang, J. Biener, R.J. Madix, J. Phys. Chem. 104 (2000) 3286. [358] M. Sambi, M.D. Negra, G. Granozzi, Z.S. Li, J.H. Jorgensen, P.J. Mùller, Appl. Surf. Sci. 142 (1999) 146. [359] M.D. Negra, M. Sambi, G. Granozzi, Surf. Sci. 436 (1999) 227. [360] M.D. Negra, M. Sambi, G. Granozzi, Surf. Sci. 461 (2000) 118. [361] G.S. Wong, D.D. Kragten, J.M. Vohs, Surf. Sci. 452 (2000) L293. [362] G.S. Wong, D.D. Kragten, J.M. Vohs, J. Phys. Chem. B 105 (2001) 1366. [363] J. Marien, T. Wagner, M. RuÈhle, Mater. Res. Soc. Symp. Proc. 441 (1997) 27. [364] J. Marien, T. Wagner, M. RuÈhle, Materialwiss. Grundlagen Symp. 7 (1997) 859. [365] J. Marien, T. Wagner, C. Duscher, A. Koch, M. RuÈhle, Surf. Sci. 446 (2000). [366] Y. Gao, S.A. Chambers, J. Mater. Res. 11 (1996) 1025. [367] Y. Gao, S.A. Chambers, Mater. Lett. 26 (1996) 217. [368] Y. Gao, Y. Liang, S.A. Chambers, Surf. Sci. 348 (1996) 17. [369] S.A. Chambers, Y. Gao, Y.J. Kim, M.A. Henderson, S. Thevuthasan, S. Wen, K.L. Merkle, Surf. Sci. 365 (1996) 625. [370] S.A. Chambers, Y. Gao, R.J. Smith, J. Vac. Sci. Technol. A 14 (1996). [371] Y. Gao, Y. Liang, S.A. Chambers, Surf. Sci. 365 (1996) 638. [372] S. Thevuthasan, N.R. Shivaparan, S.A. Chambers, Appl. Surf. Sci. 115 (1997) 381. [373] S.A. Chambers, Y. Gao, S. Thevuthasan, S. Wen, K.L. Merkle, N. Shivaparan, R.J. Smith, Mater. Res. Soc. Symp. Proc. 436 (1997) 475. [374] S.A. Chambers, Y. Gao, Y.J. Kim, Surf. Sci. Spectra 5 (1998) 211. [375] M. Casarin, C. Maccato, A. Vittadini, Phys. Chem. Chem. Phys. 1 (1999) 3793. [376] D. Morris, Y. Dou, J. Rebane, C.E.J. Mitchell, R.G. Egdell, D.S.L. Law, A. Vittadini, M. Casarin, Phys. Rev. B 61 (2000) 13445. [377] U. Diebold, J.M. Pan, T.E. Madey, Surf. Sci. 288 (1993) 896. [378] J.M. Pan, U. Diebold, L. Zhang, T.E. Madey, Surf. Sci. 295 (1993) 411. [379] J.M. Pan, B.L. Maschhoff, U. Diebold, T.E. Madey, Surf. Sci. 291 (1993) 381. [380] T.E. Madey, U. Diebold, J.M. Pan, Springer Ser. Surf. Sci. 33 (1993) 147. [381] H.A. Bullen, S.J. Garrett, Chem. Mater. 14 (2002) 243. [382] S. Petigny, B. Domenichini, H. Mostefa-Sba, E. Lesniewska, A. Steinbrunn, S. Bourgeois, Appl. Surf. Sci. 142 (1999) 114. [383] K. Asakura, W.J. Chun, Y. Iwasawa, Jpn. J. Appl. Phys. 1 (1999) 40. [384] K. Asakura, W.J. Chun, Y. Iwasawa, Shinku 42 (1999) 450. [385] W.J. Chun, K. Asakura, Y. Iwasawa, Appl. Surf. Sci. 100 (1996) 143. [386] W.J. Chun, K. Asakura, Y. Iwasawa, J. Phys. IV 2 (1997) 921. [387] W.J. Chun, K. Asakura, Y. Iwasawa, Chem. Phys. Lett. 288 (1998) 868. [388] W.J. Chun, K. Asakura, Y. Iwasawa, J. Phys. Chem. B 102 (1998) 9006. [389] W.J. Chun, K. Asakura, Y. Iwasawa, Catal. Today44 (1998) 309. [390] Y. Yamaguchi, W.J. Chun, S. Suzuki, H. Onishi, K. Asakura, Y. Iwasawa, Res. Chem. Intermed. 24 (1998) 151. [391] J. Haber, P. Nowak, J. Stoch, Bull. Pol. Acad. Sci., Chem. 45 (1997) 139. [392] V. Blondeau-Patissier, B. Domenichini, A. Steinbrunn, S. Bourgeois, Appl. Surf. Sci. 175±176 (2001) 674. U. Diebold / Surface Science Reports 48 (2003) 53±229 221

[393] S.A. Chambers, Y. Liang, Surf. Sci. 420 (1999) 123. [394] D. Brugniau, S.D. Parker, G.E. Rhead, Thin Solid Films 121 (1984) 247. [395] J. Deng, D. Wang, X. Wei, R. Zhai, H. Wang, Surf. Sci. 249 (1991) 213. [396] A.K. See, R.A. Bartynski, Phys. Rev. B 50 (1994) 12064. [397] H. Mostefa-Sba, B. Domenichini, S. Bourgeois, Surf. Sci. 437 (1999) 107. [398] R.J. Lad, V.E. Henrich, J. Vac. Sci. Technol. A 7 (1989) 1893. [399] R.J. Lad, V.E. Henrich, Phys. Rev. B 39 (1989) 13478. [400] G.A. Rizzi, A. Magrin, G. Granozzi, Surf. Sci. 443 (1999) 277. [401] G.A. Rizzi, A. Magrin, G. Granozzi, Phys. Chem. Chem. Phys. 1 (1999) 709. [402] G.A. Rizzi, M. Sambi, A. Magrin, G. Granozzi, Surf. Sci. 454±456 (2000) 30. [403] L. Atanasoska, R.T. Antanasoski, F.H. Pollak, W.E. O'Grady, Surf. Sci. 230 (1990) 95. [404] K. RavindranathanThampi, L. Lucarelli, J. Kiwi, Langmuir 7 (1991). [405] Y. Shao, W. Chen, E. Wold, J. Paul, Langmuir 10 (1994) 178. [406] S.N. Towle, G.E. Brown Jr., G.A. Parks, J. Colloid Interf. Sci. 217 (1999) 299. [407] A. BerkoÂ, G. Menesi, F. Solymosi, Surf. Sci. 372 (1997) 202. [408] A. BerkoÂ, I. Ulrych, K.C. Prince, J. Phys. Chem. B 102 (1998) 3379. [409] A. BerkoÂ, F. Solymosi, Surf. Sci. 400 (1998) 281. [410] C. Linsmeier, Surface-analytical Investigations of Supported Rhodium Model Catalysts, Ludwig-Maximilians UniversitaÈt, 1994. [411] J. Evans, B.E. Hayden, J.F.W. Mosselmans, A.J. Murray, NATO ASI Ser., Ser. C 398 (1993) 179. [412] B.E. Hayden, A. King, M.A. Newton, Surf. Sci. 397 (1998) 306. [413] G.E. Poirier, B.K. Hance, J.M. White, J. Phys. Chem. 97 (1993) 5965. [414] A. BerkoÂ, G. Klivenyi, F. Solymosi, J. Catal. 182 (1999) 511. [415] A. BerkoÂ, F. Solymosi, Surf. Sci. 411 (1998) L900. [416] C.C. Kao, S.C. Tsai, M.K. Bahl, Y.W. Chung, W.J. Lo, Surf. Sci. 95 (1980) 1. [417] H. Onishi, T. Aruga, C. Egawa, Y. Iwasawa, Surf. Sci. 233 (1990) 261. [418] M.C. Wu, P.J. Mùller, Springer Ser. Surf. Sci. 24 (1991) 652. [419] M.C. Wu, P.J. Mùller, Surf. Sci. 250 (1991) 179. [420] M.C. Wu, P.J. Mùller, Surf. Sci. 279 (1992) 23. [421] S. Bourgeois, P.L. Seigneur, M. Perdereau, D. Chandesris, P.L. Fevre, H. Magnan, Thin Solid Films 304 (1997) 267. [422] R.E. Tanner, I. Goldfarb, M.R. Castell, G.A.D. Briggs, Surf. Sci. 486 (2001) 167. [423] P.L. Cao, D.E. Ellis, V.P. Dravid, J. Mater. Res. 14 (1999) 3684. [424] S. Bourgeois, D. Diakite, F. Jomard, Surf. Sci. 217 (1989) 78. [425] S. Bourgeois, F. Jomard, M. Perderau, Surf. Sci. 249 (1991) 194. [426] S. Bourgeois, D. DiakiteÂ, F. Jomard, M. Perdereau, R. Poirault, Surf. Sci. 217 (1989) 78. [427] P. Stone, R.A. Bennett, S. Poulston, M. Bowker, Surf. Sci. 435 (1999) 501. [428] P. Stone, Chem. Commun. 13 (1998) 1369. [429] M.J.J. Jak, C. Konstapel, A. van Kreuningen, J. Verhoeven, J.W.M. Frenken, Surf. Sci. 457 (2000) 295. [430] T. Bredow, G. Pacchioni, Surf. Sci. 426 (1999) 106. [431] T. Suzuki, R. Souda, Surf. Sci. 448 (2000) 33. [432] Z. Chang, G. Thornton, Surf. Sci. 459 (2000) 303. [433] A.J. Ramirez-Cuesta, R.A. Bennett, P. Stone, P.C.H. Mitchell, M. Bowker, J. Mol. Catal. A 167 (2001) 171. [434] J. Evans, B.E. Hayden, G. Lu, Surf. Sci. 360 (1996) 61. [435] P.W. Murray, J. Shen, G. Thornton, Surf. Sci. 380 (1997). [436] S. Fischer, K.D. Schierbaum, W. GoÈpel, Vacuum 48 (1997) 601. [437] S. Gan, Y. Liang, D.R. Baer, M.R. Sievers, G.S. Herman, C.H.F. Peden, J. Phys. Chem. B 105 (2001) 2412. [438] S. Fischer, J.A. Martin-Gago, E. RomaÂn, K.D. Schierbaum, J.L. de Segovia, J. Electr. Spectrosc. Rel. Phen. 83 (1997) 217. [439] K.D. Schierbaum, X. Wei-Xing, S. Fischer, W. GoÈpel, Springer Ser. Surf. Sci. 33 (1993) 268. [440] K. Tamura, M. Kudo, M. Owari, Y. Nihei, Chem. Lett. 11 (1986) 1921. [441] R.T. Baker, E.B. Prestidge, R.L. Garten, J. Catal. 56 (1979) 390. [442] H.P. SteinruÈck, F. Pesty, L. Zhang, T.E. Madey, Phys. Rev. B 51 (1995) 2427. 222 U. Diebold / Surface Science Reports 48 (2003) 53±229

[443] W.-x. Xu, K.D. Schierbaum, W. GoÈpel, Chem. Res. Chin. Univ. 11 (1995) 219. [444] A. Linsebigler, C. Rusu, J.T. Yates Jr., J. Am. Chem. Soc. 118 (1996) 5284. [445] K.D. Schierbaum, S. Fischer, G. Thornton, Surf. Sci. 391 (1997) 196. [446] S. Gan, A. El-azab, Y. Liang, Surf. Sci. 479 (2001) L369. [447] U. Diebold, J.M. Pan, T.E. Madey, Phys. Rev. B 47 (1993) 3868. [448] D.L. Carroll, M. Wagner, M. RuÈhle, D.A. Bonnell, J. Mater. Res. 12 (1997) 975. [449] P.J. Mùller, M.C. Wu, Surf. Sci. 224 (1989) 265. [450] P. Lu, F. Cosandey, Interf. Sci. 2 (1994) 169. [451] Z. Chang, S. Haq, G. Thornton, Surf. Sci. 467 (2000) L841. [452] P.L. Wincott, J.S.G. Irwin, G. Jones, P.J. Hardman, G. Thornton, S. Weichel, P.J. Mùller, V.R. Dhanak, Surf. Sci. 379 (1997) 242. [453] K. Luo, T.P. St. Clair, X. Lai, D.W. Goodman, J. Phys. Chem. 104 (2000) 3050. [454] D. Martin, J. Jupille, Y. Borensztein, Surf. Sci. 404 (1998) 433. [455] D. Martin, F. Creuzet, J. Jupille, Y. Borensztein, P. Gadenne, Surf. Sci. 379 (1997) 958. [456] D.A. Chen, M.C. Bartelt, K.F. McCarty, Surf. Sci. 464 (2000) L708. [457] C. Su, J.C. Yeh, J.L. Lin, J.C. Lin, Appl. Surf. Sci. 169±170 (2001) 366. [458] M. Haruta, Catal. Today36 (1997) 153. [459] M. Haruta, S. Tsubota, T. Kobayashi, H. Kageyama, M.J. Genet, B. Delmont, J. Catal. 144 (1993) 175. [460] G.C. Bond, D.T. Thompson, Catal. Rev.-Sci. Eng. 418 (1999) 319. [461] M. Valden, S. Pak, X. Lai, D.W. Goodman, Catal. Lett. 56 (1998) 7. [462] J.-D. Grunwaldt, A. Baiker, J. Phys. Chem. B 103 (1999) 1002. [463] A. Sanchez, S. Abbet, U. Heiz, W.-D. Schneider, H. HaÈkkinen, R. N. Barnett, U. Landmann, J. Phys. Chem. A 103 (1999). [464] F. Cosandey, L. Zhang, T.E. Madey, Surf. Sci. 474 (2001) 1. [465] L. Zhang, R. Persaud, T.E. Madey, Phys. Rev. B 56 (1997) 10549. [466] T. Akita, K. Tanaka, S. Tsubota, M. Haruta, J. Electr. Microsc. 49 (2000) 657. [467] X. Lai, T.P. St. Clair, D.W. Goodman, Prog. Surf. Sci. 59 (1998) 25. [468] M. Okumura, Y. Kitagawa, M. Haruta, K. Yamaguchi, Chem. Phys. Lett. 346 (2001) 163. [469] J.A. Rodriguez, G. Liu, T. Jirsak, J. Hrbek, Z. Chang, J. Dvorak, A. Maiti, J. Am. Chem. Soc. 124 (2002) 5242. [470] H. Kobayashi, K. Kishimoto, Y. Nakato, H. Tsubomura, Sens. Actuators B 13 (1993) 125. [471] P. Salvador, M.L. Garcia Gonzalez, F. Munoz, J. Phys. Chem. 96 (1992) 10349. [472] V.E. Henrich, R.L. Kurtz, Phys. Rev. B 23 (1981) 6280. [473] M.S. Lazarus, T.K. Sham, Chem. Phys. Lett. 92 (1982) 670. [474] Q. Zhong, J.M. Vohs, D.A. Bonnell, J. Am. Ceram. Soc. 76 (1993) 1137. [475] W.E. Wallace, Q. Zhong, J. Genzer, R.J. Composto, D.A. Bonnell, J. Mater. Res. 8 (1993) 1629. [476] T. Fujino, M. Katayama, K. Inudzuka, T. Okuno, K. Oura, T. Hirao, Appl. Phys. Lett. 79 (2001) 2716. [477] M.A. Henderson, Surf. Sci. Rep. 46 (2002) 1. [478] P.A. Thiel, T.E. Madey, Surf. Sci. Rep. 7 (1987) 211. [479] C.A. Muryn, G. Tirvengadum, J.J. Crouch, D.R. Warburton, G.N. Raiker, G. Thornton, D.S.L. Law, Phys. Rev. B 1 (1989) SB127. [480] G. Thornton, Springer Ser. Surf. Sci. 33 (1993) 115. [481] C.A. Muryn, P.J. Hardman, J.J. Crouch, G.N. Raiker, G. Thornton, D.S.L. Law, Surf. Sci. 251±252 (1991) 747. [482] S. Bourgeois, F. Jomard, M. Perdereau, Surf. Sci. 279 (1992) 349. [483] S. Bourgeois, L. Gitton, M. Perdereau, J. Chim. Phys. Phys.-Chim. Biol. 85 (1988) 413. [484] L.Q. Wang, P.X. Skiba, A.N. Shultz, D.R. Baer, M.H. Engelhard, Mater. Res. Soc. Symp. Proc. 432 (1997) 45. [485] R. Wang, N. Sakai, A. Fujishima, T. Watanabe, K. Hashimoto, J. Phys. Chem. B 103 (1999) 2188. [486] R.F. Nalewajski, A.M. Koester, T. Bredow, K. Jug, J. Mol. Catal. 82 (1993) 407. [487] D. Brinkley, M. Dietrich, T. Engel, P. Farrall, G. Gantner, A. Schafer, A. Szuchmacher, Surf. Sci. 395 (1998) 292. [488] I.M. Brookes, C.A. Muryn, G. Thornton, Phys. Rev. Lett. 87 (2001) 266103/1. [489] R.L. Kurtz, R. Stockbauer, T.E. Madey, E. RomaÂn, J.L. de Segovia, Surf. Sci. 218 (1989) 178. [490] G. Lu, A. Linsebigler, J.T. Yates Jr., J. Phys. Chem. 98 (1994) 11733. [491] S. Suzuki, K.-i. Fukui, H. Onishi, T. Sasaki, Y. Iwasawa, Stud. Surf. Sci. Catal. 132 (2001) 753. [492] P.J.D. Lindan, N.M. Harrison, J.M. Holender, M.J. Gillan, Chem. Phys. Lett. 261 (1996) 246. U. Diebold / Surface Science Reports 48 (2003) 53±229 223

[493] P.J.D. Lindan, N.M. Harrison, M.J. Gillan, Phys. Rev. Lett. 80 (1998) 762. [494] E.V. Stefanovich, T.N. Truong, Chem. Phys. Lett. 299 (1999) 623. [495] G.S. Herman, Z. Dohnalek, N. Ruzycki, U. Diebold, J. Phys. Chem. B, submitted. [496] G. Lu, A. Linsebigler, J.T. Yates Jr., J. Chem. Phys. 102 (1995) 4657. [497] G. Lu, A. Linsebigler, J.T. Yates Jr., J. Chem. Phys. 102 (1995) 3005. [498] C.N. Rusu, J.T. Yates Jr., Langmuir 13 (1997) 4311. [499] W.S. Epling, C.H.F. Peden, M.A. Henderson, U. Diebold, Surf. Sci. 412 (1998) 333. [500] C.L. Perkins, M.A. Henderson, J. Phys. Chem. B 105 (2001) 3856. [501] A.N. Shultz, W.M. Hetherington, Iii, D.R. Baer, L.Q. Wang, M.H. Engelhard, Surf. Sci. 392 (1997) 1. [502] X.-G. Wang, W. Weiss, S. Shaikhutdinov, M. Ritter, M. Petersen, F. Wagner, R. SchloÈgl, M. Scheffler, Phys. Rev. Lett. 81 (1998) 1038. [503] X.-G. Wang, A.M. Chaka, M. Scheffler, Phys. Rev. Lett. 84 (2000) 3659. [504] K. Reuter, M. Scheffler, Phys. Rev. B 65 (2001) 035406. [505] A. Boffa, C. Lin, A.T. Bell, G.A. Somorjai, J. Catal. 149 (1994) 149. [506] M.C. Torquemada, J.L. de Segovia, Surf. Sci. 331±333 (1995) 219. [507] M.-C. Wu, P.J. Mùller, Surf. Sci. 250 (1991) 179. [508] A. BerkoÂ, G. Menesi, F. Solymosi, J. Phys. Chem. 100 (1996) 17732. [509] D.W. Goodman, T.P.S. Clair, X. Lai, M. Valden, Morphologyand electronic structure of metal clusters supported on

TiO2(1 1 0), in: Book of Abstracts, Proceedings of the 217th ACS National Meeting, Anaheim, CA, March 1999. [510] A. Linsebigler, G. Lu, J.T. Yates Jr., J. Chem. Phys. 103 (1995) 9438. [511] G. Rocker, W. GoÈpel, Surf. Sci. 175 (1986) L675. [512] M.C. Torquemada, J.L. de Segovia, E. RomaÂn, Surf. Sci. 337 (1995) 31. [513] D.C. Sorescu, J.T. Yates Jr., J. Phys. Chem. B 102 (1998) 4556. [514] H. Kobayashi, M. Yamaguchi, Surf. Sci. 214 (1989) 466. [515] A. Fahmi, C. Minot, Organometall. Chem. 478 (1994) 67. [516] Z. Yang, R. Wu, Q. Zhang, D.W. Goodman, Phys. Rev. B 63 (2001) 045419/1. [517] G. Pacchioni, A.M. Ferrari, P.S. Bagus, Surf. Sci. 350 (1996) 159. [518] M. Casarin, C. Maccato, A. Vittadini, Appl. Surf. Sci. 142 (1999) 196. [519] M. Casarin, C. Maccato, A. Vittadini, J. Phys. Chem. B 103 (1999) 3510. [520] P. Reinhardt, M. Causa, C.M. Marian, B.A. Heû, Phys. Rev. B 54 (1996) 14812. [521] G.B. Raupp, J.A. Dumesic, J. Phys. Chem. 89 (1985) 5240. [522] A. Linsebigler, G. Lu, J.T. Yates Jr., J. Phys. Chem. 100 (1996) 6631. [523] K. Tanaka, J.M. White, J. Catal. 79 (1983) 81. [524] H. Yamashita, N. Kamada, H. He, K. Tanaka, S. Ehara, M. Anpo, Chem. Lett. 5 (1994) 855. [525] R.I. Bickley, R.K.M. Jayanty, J.A. Navio, C. Real, M. Macias, Surf. Sci. 251±252 (1991) 1052. [526] F. Rittner, B. Boddenberg, M.J. Bojan, W.A. Steele, Langmuir 15 (1999) 1456. [527] F. Rittner, B. Boddenberg, R.F. Fink, V. Staemmler, Langmuir 15 (1999) 1449. [528] S. Wang, B. Raman, D.H. Chen, K.Y. Li, J.A. Colapret, Chem. Oxid. 5 (1997) 289. [529] D.C. Sorescu, C.N. Rusu, J.T. Yates Jr., J. Phys. Chem. B 104 (2000) 4408. [530] C.N. Rusu, J.T. Yates Jr., J. Phys. Chem. B 105 (2001) 2596. [531] C.N. Rusu, J.T. Yates Jr., J. Phys. Chem. 104 (2000) 1729. [532] L.Q. Wang, A.N. Shultz, D.R. Baer, M.H. Engelhard, J. Vac. Sci. Technol. A 14 (1996) 1532. [533] J.A. Rodriguez, T. Jirsak, G. Liu, J. Hrbek, J. Dvorak, A. Maiti, J. Am. Chem. Soc. 123 (2001) 9597. [534] U. Diebold, T.E. Madey, J. Vac. Sci. Technol. A 10 (1992) 2327. [535] U. Diebold, T.E. Madey, Springer Ser. Surf. Sci. 31 (1993) 284. [536] E. RomaÂn, J.L. de Segovia, Surf. Sci. 251/252 (1991) 742. [537] E.L. RomaÂn, J.L. de Segovia, R.L. Kurtz, T.E. Madey, Surf. Sci. 273 (1992) 40. [538] A. Markovits, J. Ahdjoudj, C. Minot, Surf. Sci. 365 (1996) 649. [539] D. Paschek, A. Geiger, AIP Conf. Proc. 330 (1995) 349. [540] W.K. Siu, R.A. Bartynski, S.L. Hulbert, J. Chem. Phys. 113 (2000) 10697. [541] D.R. Warburton, D. Purdie, C.A. Muryn, K. Prabhakaran, P.L. Wincott, G. Thornton, Surf. Sci. 269/270 (1992) 305. 224 U. Diebold / Surface Science Reports 48 (2003) 53±229

[542] C.A. Muryn, D. Purdie, P. Hardman, A.L. Johnson, N.S. Prakash, G.N. Raiker, G. Thornton, D.S.L. Law, Faraday Discuss., Chem. Soc. 89 (1990) 77. [543] K.E. Smith, J.L. Mackay, V.E. Henrich, Phys. Rev. B 35 (1987) 5822. [544] K.E. Smith, V.E. Henrich, J. Vac. Sci. Technol. A 7 (1989) 1967. [545] H. Onishi, T. Aruga, C. Egawa, Y. Iwasawa, Surf. Sci. 193 (1988) 33. [546] C.L. Greenwood, E.M. Williams, G. Thornton, S.L. Bennett, E. RomaÂn, J.L. de Segovia, M.C. Torquemada, Surf. Sci. 288 (1993) 386. [547] M.C. Torquemada, J.L. de Segovia, E. RomaÂn, G. Thornton, E.M. Williams, S.L. Bennett, Springer Ser. Surf. Sci. 31 (1993) 289. [548] M.C. Torquemada, J.L. de Segovia, J. Vac. Sci. Technol. A 12 (1994) 2318. [549] J.L. de Segovia, M.C. Torquemada, E. RomaÂn, J. Phys.: Cond. Matt. 5 (1993) A 139. [550] H. Raza, S.P. Harte, A. Rodriguez, Surf. Sci. 366 (1996) 519. [551] M.A. Barteau, in: D.P. Woodruff (Ed.), The Chemical Physics of Solid Surfaces, vol. 9, Oxide Surfaces, Elsevier, Amsterdam, 2001. [552] K.E. Smith, V.E. Henrich, Surf. Sci. 217 (1989) 445. [553] A. Selloni, A. Vittadini, M. GraÈtzel, Surf. Sci. 402±404 (1998) 219. [554] W. Heegemann, K.H. Meister, E. Bechtold, K. Hayek, Surf. Sci. 49 (1975) 161. [555] E.L.D. Hebenstreit, W. Hebenstreit, U. Diebold, Surf. Sci. 470 (2001) 347. [556] E.L.D. Hebenstreit, W. Hebenstreit, H. Geisler, S.N. Thornburg, C.A. Ventrice Jr., D.A. Hite, P.T. Sprunger, U. Diebold, Phys. Rev. B 64 (2001) 115418/1. [557] J. Hrbek, J.A. Rodriguez, T. Jirsak, J. Dvorak, J. Electr. Spectrosc. Rel. Phen. 119 (2001) 201. [558] J. Hrbek, J.A. Rodriguez, J. Dvorak, T. Jirsak, Collect. Czech. Chem. Commun. 66 (2001) 1149. [559] E.L.D. Hebenstreit, W. Hebenstreit, H. Geisler, C.A. Ventrice Jr., P.T. Sprunger, U. Diebold, Surf. Sci. 486 (2001) L467. [560] U. Diebold. Understanding metal oxide surfaces at the atomic scale: STM investigations of bulk-defect dependent surface processes, in: X.P.C.B. Carter, K. Sickafus, H.L. Tuller, T. Wood (Eds.), Material Research SocietySymposium Proceedings, vol. 654, Materials Research Society, 2001. [561] E.L.D. Hebenstreit, W. Hebenstreit, C.A. Ventrice Jr., H. Geisler, D.A. Hite, P.T. Sprunger, U. Diebold, Surf. Sci. 505 (2002) 336. [562] D. Vogtenhuber, R. Podloucky, J. Redinger, Surf. Sci. 454±456 (2000) 369. [563] H. Brune, J. Wintterlin, R.J. Behm, G. Ertl, Phys. Rev. Lett. 68 (1992) 624. [564] C. Engdahl, G. WahnstroÈm, Surf. Sci. 312 (1994) 429. [565] B. Kasemo, E. ToÈrnqvist, J.K. Nùrskov, B.I. Lundqvist, Surf. Sci. 89 (1979) 554. [566] G. WahnstroÈm, A.B. Lee, J. StroÈmquist, J. Chem. Phys. 105 (1996) 326. [567] J. Wintterlin, R. Schuster, G. Ertl, Phys. Rev. Lett. 77 (1996) 123. [568] J.A. Jensen, C. Yan, A.C. Kummel, Phys. Rev. Lett. 76 (1996) 1388. [569] Y.L. Li, D.P. Pullman, J.J. Yang, A.A. Tsekouras, D.B. Gosalvez, K.B. Laughlin, Z. Zhang, M.T. Schulberg, D.J. Gladstone, M. McGonigal, S.T. Ceyer, Phys. Rev. Lett. 74 (1995) 2603. [570] M. Alam, M.A. Henderson, P.D. Kaviratna, G.S. Herman, C.H.F. Peden, J. Phys. Chem. B 102 (1998) 111. [571] P. Dolle, K. Markert, W. Heichler, N.R. Armstrong, K. Wandelt, K.S. Kim, R.A. Fiato, J. Vac. Sci. Technol. A 4 (1986) 1465. [572] V.A. Bakaev, W.A. Steele, Langmuir 8 (1992) 1372. [573] F. Rittner, D. Paschek, B. Boddenberg, Langmuir 11 (1995) 3097. [574] W.A. Steele, Fortschr.Ber. VDI, Reihe 555 (1998) 47. [575] P. Jakob, M. Gsell, D. Menzel, J. Chem. Phys. 114 (2001) 10075. [576] M. Gsell, P. Jakob, D. Menzel, Science 280 (1998) 717. [577] H. Idriss, M.A. Barteau, Adv. Catal. 45 (2000) 261. [578] H. Idriss, M. Libby, M.A. Barteau, Catal. Lett. 15 (1992) 13. [579] M.A. Barteau, Stud. Surf. Sci., Catalysis A 130 (2000) 105. [580] M.A. Barteau, Chem. Rev. 96 (1996) 1413. [581] H. Onishi, T. Aruga, Y. Iwasawa, J. Catal. 146 (1994) 557. [582] M.A. Henderson, J. Phys. Chem. B 101 (1997) 221. [583] H. Idriss, V.S. Lusvardi, M.A. Barteau, Surf. Sci. 348 (1996) 39. U. Diebold / Surface Science Reports 48 (2003) 53±229 225

[584] R.A. Bennett, P. Stone, R. Smith, M. Bowker, Surf. Sci. 454±456 (2000) 390. [585] S.A. Chambers, M.A. Henderson, Y.J. Kim, S. Thevuthasan, Surf. Rev. Lett. 5 (1998) 381. [586] S.A. Chambers, S. Thevuthasan, Y.J. Kim, G.S. Herman, Z. Wang, E. Tober, R. Ynzunza, J. Morais, C.H.F. Peden, K. Ferris, C.S. Fadley, Chem. Phys. Lett. 267 (1997) 51. [587] S. Thevuthasan, G.S. Herman, Y.J. Kim, S.A. Chambers, C.H.F. Peden, Z. Wang, R.X. Ynzunza, E.D. Tober, J. Morais, C.S. Fadley, Surf. Sci. 401 (1998) 261. [588] B.E. Hayden, A. King, M.A. Newton, J. Phys. Chem. B 103 (1999) 203. [589] A. Gutierrez-Sosa, P. Martinez-Escolano, H. Raza, R. Lindsay, P.L. Wincott, G. Thornton, Surf. Sci. 471 (2001) 163. [590] P. KaÈckell, K. Terakura, Appl. Surf. Sci. 166 (2000) 370. [591] P. KaÈckell, K. Terakura, Surf. Sci. 461 (2000) 191. [592] J. Ahdjoudj, C. Minot, Catal. Lett. 46 (1997) 83. [593] L.Q. Wang, K.F. Ferris, A.N. Schultz, D.R. Baer, M.H. Engelhard, Surf. Sci. 380 (1997) 352. [594] K.-i. Fukui, H. Onishi, Y. Iwasawa, Chem. Phys. Lett. 280 (1997) 296. [595] K.-i. Fukui, H. Onishi, Y. Iwasawa, Appl. Surf. Sci. 140 (1999) 259. [596] H. Onishi, Y. Iwasawa, Jpn. J. Appl. Phys. 33 (1994) L1338. [597] H. Onishi, K.-i. Fukui, Y. Iwasawa, Colloids Surf. A 109 (1996) 335. [598] H. Onishi, Y. Iwasawa, Surf. Sci. 358 (1996) 773. [599] H. Onishi, K.-i. Fukui, Y. Iwasawa, Jpn. J. Appl. Phys. 38 (1999) 3830. [600] A. Sasahara, H. Uetsuka, H. Onishi, J. Phys. Chem. B 105 (2001) 1. [601] Z. Chang, G. Thornton, Surf. Sci. 462 (2000) 68. [602] A. Atrei, U. Bardi, G. Rovida, Surf. Sci. 391 (1997) 216. [603] K.S. Kim, M.A. Barteau, Langmuir 6 (1990) 1485. [604] K.S. Kim, M.A. Barteau, J. Catal. 125 (1990) 353. [605] A. Vittadini, A. Selloni, M. GraÈtzel, J. Phys. Chem. 104 (2000) 1300. [606] H. Onishi, T. Aruga, Y. Iwasawa, J. Am. Chem. Soc. 115 (1993) 10460. [607] A. Ludviksson, R. Zhang, C.T. Campbell, K. Griffiths, Surf. Sci. 313 (1994) 64. [608] I.D. Cocks, Q. Guo, R. Patel, E.M. Williams, E. RomaÂn, J.L. de Segovia, Surf. Sci. 379 (1997) 135. [609] H. Onishi, Y. Yamaguchi, K.-i. Fukui, Y. Iwasawa, J. Phys. Chem. 100 (1996) 9582. [610] A. Sasahara, H. Uetsuka, H. Onishi, Surf. Sci. 481 (2001) L437. [611] K.-i. Fukui, Y. Iwasawa, Surf. Sci. 464 (2000) L719. [612] D.J. Titheridge, M.A. Barteau, H. Idriss, Langmuir 17 (2001) 2120. [613] Q. Guo, I. Cocks, E.M. Williams, Surf. Sci. 393 (1997) 1. [614] Q. Guo, E.M. Williams, Surf. Sci. 435 (1999) 322. [615] L. Patthey, H. Rensmo, P. Persson, K. Westermark, L. Vayssieres, A. Stashans, A. Petersson, P.A. BruÈhwiler, H. Siegbahn, S. Lunell, N. MaÊrtensson, J. Chem. Phys. 110 (1999) 5913. [616] P. Persson, S. Lunell, P.A. Bruhwiler, J. Schnadt, S. Sodergren, J.N. O'Shea, O. Karis, H. Siegbahn, N. MaÊrtensson, M. Bassler, L. Patthey, J. Chem. Phys. 112 (2000) 3945. [617] A. Fahmi, C. Minot, P. Fourre, P. Nortier, Surf. Sci. 343 (1995) 261. [618] E. Soria, E. RomaÂn, E.M. Williams, J.L. de Segovia, Surf. Sci. 435 (1999) 543. [619] E. Soria, I. Colera, J.L. de Segovia, Surf. Sci. 451 (2000) 188. [620] J.N. Wilson, D.J. Titheridge, L. Kieu, H. Idriss, J. Vac. Sci. Technol. A 18 (2000) 1887. [621] M.A. Henderson, S. Otero-Tapia, M.E. Castro, FaradayDiscuss., Chem. Soc. 114 (1999) 313. [622] L.Q. Wang, K.F. Ferris, J.P. Winokur, A.N. Shultz, D.R. Baer, M.H. Engelhard, J. Vac. Sci. Technol. A 16 (1998) 3034. [623] S.P. Bates, M.J. Gillan, G. Kresse, J. Phys. Chem. B 102 (1998) 2017. [624] G. Kresse, J. Furthmueller, Phys. Rev. B 54 (1996) 11169. [625] G. Kresse, J. Furthmuller, Comput. Mater. Sci. 6 (1996) 15. [626] G. Kresse, J. Hafner, Phys. Rev. B 47 (1993) 558. [627] K.S. Kim, M.A. Barteau, Surf. Sci. 223 (1989) 13. [628] E. RomaÂn, F.J. Bustillo, J.L. de Segovia, Vacuum 41 (1990) 40. [629] K.S. Kim, M.A. Barteau, J. Mol. Catal. 63 (1990) 103. [630] L. Gamble, L.S. Jung, C.T. Campbell, Surf. Sci. 348 (1996) 1. [631] V.S. Lusvardi, M.A. Barteau, W.R. Dolinger, W.E. Farneth, J. Phys. Chem. 100 (1996) 18183. 226 U. Diebold / Surface Science Reports 48 (2003) 53±229

[632] H. Idriss, K.S. Kim, M.A. Barteau, Surf. Sci. 262 (1992) 113. [633] H. Idriss, K.S. Kim, M.A. Barteau, J. Catal. 139 (1993) 119. [634] H. Idriss, M.A. Barteau, Catal. Lett. 40 (1996) 147. [635] H. Idriss, K.G. Pierce, M.A. Barteau, J. Am. Chem. Soc. 116 (1994) 3063. [636] A.B. Sherrill, V.S. Lusvardi, J. Eng, J.G. Chen, M.A. Barteau, Catal. Today63 (2000) 43. [637] S. Cannizzaro, Ann. Chemie Pharmac. 88 (1853) 129. [638] K.G. Pierce, M.A. Barteau, J. Org. Chem. 60 (1995) 2405. [639] H. Idriss, M.A. Barteau, Langmuir 10 (1994) 3693. [640] H. Idriss, M.A. Barteau, Reductive coupling of cyclic ketones on reduced titania (0 0 1) single crystal surfaces, in: M. Guisnet (Ed.), Heterogeneous Catalysis and Fine Chemicals III, vol. 78, Elsevier, Amsterdam, 1993, 463 pp. [641] K.G. Pierce, M.A. Barteau, J. Phys. Chem. 98 (1994) 3882. [642] K.G. Pierce, V.S. Lusvardi, M.A. Barteau, Catalytic formation of carbon±carbon bonds in ultrahigh vacuum:

cyclotrimerization of alkynes on reduced TiO2 surfaces, in: J.W. Hightower, W.N. Delgass, E. Iglesia, A.T. Bell (Eds.), Proceedings of the 11th International Congress on CatalysisÐ40th Anniversary, Studies in Surface Science and Catalysis, vol. 101, Elsevier, Amsterdam, 1996. [643] K.G. Pierce, M.A. Barteau, Surf. Sci. 326 (1995) L473. [644] A.B. Sherrill, M.A. Barteau, J. Mol. Catal. A 184 (2002) 301. [645] K.G. Pierce, M.A. Barteau, J. Mol. Catal. 94 (1994) 389. [646] A.B. Sherrill, J.W. Medlin, J.G. Shen, M.A. Barteau, Surf. Sci. (2001). [647] A.B. Sherrill, V.S. Lusvardi, M.A. Barteau, Langmuir 15 (1999) 7615. [648] K. Sakamaki, K. Itoh, A. Fujishima, Y. Gohshi, J. Vac. Sci. Technol. A 8 (1990) 614. [649] K. Sakamaki, S. Matsunaga, K. Itoh, A. Fujishima, Y. Gohshi, Surf. Sci. 219 (1989) L531. [650] S. Suzuki, Y. Yamaguchi, H. Onishi, T. Sasaki, K.-i. Fukui, Y. Iwasawa, J. Chem. Soc., FaradayTrans. 94 (1998) 161. [651] H. Raza, P.L. Wincott, G. Thornton, R. Casanova, A. Rodriguez, Surf. Sci. 402±404 (1998) 710. [652] S. Suzuki, H. Onishi, K.-i. Fukui, Y. Iwasawa, Chem. Phys. Lett. 304 (1999) 225. [653] T. Sasaki, K.-i. Fukui, Y. Iwasawa, Stud. Surf. Sci. Catal. 132 (2001) 749. [654] S. Suzuki, H. Onishi, T. Sasaki, K.-i. Fukui, Y. Iwasawa, Catal. Lett. 54 (1998) 177. [655] L. Gamble, L.S. Jung, C.T. Campbell, Langmuir 11 (1995) 4505. [656] L. Gamble, M.B. Hugenschmidt, C.T. Campbell, T.A. Jurgens, J.W. Rogers Jr., J. Am. Chem. Soc. 115 (1993) 12096. [657] L. Gamble, M.A. Henderson, C.T. Campbell, J. Phys. Chem. B 102 (1998) 4536. [658] M.A. Fox, M.T. Dulay, Chem. Rev. 93 (1993) 341. [659] A. Hagfeldt, M. GraÈtzel, Chem. Rev. (Washington, DC) 95 (1995) 49. [660] E. Wilson, C&EN 1 (1996) 29. [661] A. Hagfeldt, M. GraÈtzel, Acc. Chem. Res. 33 (2000) 269. [662] V. Shklover, M. GraÈtzel, Proc. SPIE-Int. Soc. Opt. Eng. 3469 (1998) 134. [663] K.S. Finnie, J.R. Bartlett, J.L. Woolfrey, Langmuir 14 (1998) 2744. [664] G. Lu, A. Linsebigler, J.T. Yates Jr., J. Phys. Chem. 99 (1995) 7626. [665] J.C.S. Wong, A. Linsebigler, G. Lu, J. Fan, J.T. Yates Jr., J. Phys. Chem. 99 (1995) 335. [666] D. Brinkley, T. Engel, Surf. Sci. 415 (1998) L1001. [667] D. Brinkley, T. Engel, J. Phys. Chem. B 102 (1998) 7596. [668] D. Brinkley, T. Engel, J. Phys. Chem. B 104 (2000) 9836. [669] V. Shapovalov, E.V. Stefanovich, T.N. Truong, Surf. Sci. 498 (2002) L103. [670] D.H. Fairbrother, V.P. Holbert, K.A. Briggman, P.C. Stair, E. Weitz, Proc. SPIE Int. Soc. Opt. Eng. 2547 (1995) 248. [671] S.J. Garrett, V.P. Holbert, P.C. Stair, E. Weitz, J. Chem. Phys. 100 (1994) 4615. [672] S.J. Garrett, V.P. Holbert, P.C. Stair, E. Weitz, J. Chem. Phys. 100 (1994) 4626. [673] V.P. Holbert, S.J. Garrett, P.C. Stair, E. Weitz, Surf. Sci. 346 (1996) 189. [674] S.H. Kim, K.A. Briggman, P.C. Stair, E. Weitz, J. Vac. Sci.Technol. A 14 (1996). [675] S.H. Kim, P.C. Stair, E. Weitz, Langmuir 14 (1998) 4156. [676] S.H. Kim, P.C. Stair, E. Weitz, J. Chem. Phys. 108 (1998) 5080. [677] S.H. Kim, P.C. Stair, E. Weitz, Chem. Phys. Lett. 302 (1999) 511. [678] S.H. Kim, P.C. Stair, E. Weitz, Surf. Sci. 445 (2000) 177. U. Diebold / Surface Science Reports 48 (2003) 53±229 227

[679] P. Antoniewicz, Phys. Rev. B 21 (1980) 3811. [680] M.L. Garcia Gonzalez, P. Salvador, J. Electroanal. Chem. 325 (1992) 369. [681] M.L. Garcia Gonzalez, P. Salvador, J. Electroanal. Chem. 326 (1992) 323. [682] M.L. Garcia Gonzalez, A. Martinez Chaparro, P. Salvador, J. Photochem. Photobiol. A 73 (1993) 221. [683] D. Tafalla, M. Pujadas, P. Salvador, Surf. Sci. 215 (1989) 190. [684] P. Salvador, Surf. Sci. 192 (1987) 36. [685] D. Tafalla, P. Salvador, Ber. Bunsenges. Phys. Chem. 91 (1987) 475. [686] M.T. Spitler, M. Calvin, J. Chem. Phys. 66 (1977) 4294. [687] J.M. Lantz, R. Baba, R.M. Corn, J. Phys. Chem. 97 (1993) 7392. [688] J.M. Lantz, R.M. Corn, J. Phys. Chem. 98 (1994) 9387. [689] P.R. Unwin, A.J. Bard, J. Phys. Chem. 96 (1992) 5035. [690] C.M. Wang, T.E. Mallouk, J. Phys. Chem. 94 (1990) 4276. [691] K. Shaw, P. Christensen, A. Hamnett, Electrochim. Acta 41 (1996) 719. [692] H. Yanagi, S. Chen, P.A. Lee, K.W. Nebesny, N.R. Armstrong, A. Fujishima, J. Phys. Chem. 100 (1996) 5447. [693] S.E. Gilbert, J.H. Kennedy, Surf. Sci. 225 (1990) L1. [694] S.E. Gilbert, J.H. Kennedy, J. Electrochem. Soc. 135 (1988) 2385. [695] F.R.F. Fan, A.J. Bard, J. Phys. Chem. 94 (1990) 3761. [696] D.F. Cox, T.B. Fryberger, S. Semancik, Surf. Sci. 224 (1989) 121. [697] F.H. Jones, R. Dixon, J.S. Foord, R.G. Egdell, J.B. Pethica, Surf. Sci. 367 (1997) 376. [698] C.L. Pang, S.A. Haycock, H. Raza, P.J. Mùller, G. Thornton, Phys. Rev. B 62 (2000) R7778. [699] M. Sinner-Hettenbach, M. GoÈthelid, J. Weissenrieder, H. von Schenck, T. Weiû, N. Barsan, U. Weimar, Surf. Sci. 477 (2001) 2001. [700] I. Manassidis, J. Goniakowski, L.N. Kantorovich, M.J. Gillan, Surf. Sci. 339 (1995) 258. [701] A. Atrei, E. Zanazzi, U. Bardi, G. Rovida, Surf. Sci. 475 (2001) L223. [702] H. Over, Prog. Surf. Sci. 58 (1998) 249. [703] H. Over, Y.D. Kim, A.P. Seitsonen, S. Wendt, E. Lundgren, M. Schmid, P. Varga, A. Morgante, G. Ertl, Science 287 (2000). [704] Y.D. Kim, A.P. Seitsonen, S. Wendt, J. Wang, C. Fan, K. Jacobi, H. Over, G. Ertl, J. Phys. Chem. B 105 (2001) 3752. [705] H. Over, A.P. Seitsonen, E. Lundgren, M. Schmid, P. Varga, Surf. Sci. 515 (2002) 143. [706] S.B. Basame, D. Habel-Rodriguez, D.J. Keller, Appl. Surf. Sci. 183 (2001) 62. [707] M. Kawasaki, K. Takahashi, T. Maeda, R. Tsuchiya, M. Shinohara, O. Ishiyama, T. Yonezawa, M. Yoshimoto, H. Koinuma, Science 266 (1994) 1540. [708] M.R. Castell, Surf. Sci. 516 (2002) 33. [709] N. Erdman, K.R. Poepelmeier, M. Asta, O. Warschkow, D.E. Ellis, L.D. Marks, Nature 419 (2002) 55. [710] W. Weiss, W. Ranke, Prog. Surf. Sci. 70 (2002) 1. [711] A.E. Taverner, A. Gulino, R.G. Egdell, T.J. Tate, Appl. Surf. Sci. 90 (1995) 383. [712] M. Batzill, E.L.D. Hebenstreit, W. Hebenstreit, U. Diebold, Chem. Phys. Lett., in press. [713] P. Hoyer, Langmuir 12 (1996) 1411. [714] S.M. Liu, L.M. Gan, L.H. Liu, W.D. Zhang, H.C. Zheng, Chem. Mater. 14 (2002) 1391. [715] S.D. Burnside, V. Shklover, C. Barbe, P. Comte, F. Arendse, K. Brookes, M. GraÈtzel, Chem. Mater. 10 (1998) 2419. [716] J.E. Sunstrom IV, W.R. Moser, B. Marhik-Guerts, Chem. Mater. 8 (1996). [717] H.A. Bullen, S.J. Garrett, Nano Lett. 2 (2002) 739. [718] K. Subramanya Mayya, D.I. Gittins, F. Caruso, Chem. Mater. 13 (2001) 3833. [719] F. SchuÈth, Chem. Mater. 13 (2001) 3184. [720] Y.C. Zhu, C.X. Ding, Nanostruct. Mater. 11 (1999) 427. [721] M. Iida, T. Sasaki, M. Watanabe, Chem. Mater. 10 (1998) 3780. [722] Y. Yin, Y. Lu, B. Gates, Y. Xia, Chem. Mater. 13 (2001) 1146. [723] P. Yang, D. Zhao, D.I. Margolese, B.F. Chmelka, G.D. Stucky, Nature 396 (1998) 152. [724] T. Moritz, J. Reiss, K. Diesner, D. Su, A. Chemseddine, J. Phys. Chem. B 101 (1997) 8052. [725] E. Asari, R. Souda, Surf. Sci. 486 (2001) 203. 228 U. Diebold / Surface Science Reports 48 (2003) 53±229

[726] H.-J. Freund, H. Kuhlenbeck, J. Libuda, G. Rupprechter, M. BaÈumer, H. Hamann, Top. Catal. 15 (2001) 201. [727] D.F. Ogletree, H. Bluhm, G. Lebedev, C.S. Fadley, Z. Hussain, M. Salmeron, Rev. Sci. Instrum. 73 (2002) 3872. [728] P.B. Rasmussen, B.L.M. Hendriksen, H. Zeijlemaker, H.G. Ficke, J.W.M. Frenken, Rev. Sci. Instrum. 69 (1988) 3879. [729] B.L.M. Hendriksen, J.W.M. Frenken, Phys. Rev. Lett. 89 (2002) 046101. [730] K.Y. Kung, P. Chen, F. Wei, G. Rupprechter, Y.R. Shen, G.A. Somorjai, Rev. Sci. Instrum. 72 (2001) 1806. [731] P.A.M. Hotsenpiller, J.D. Bolt, W.E. Farneth, J.B. Lowekamp, G.S. Rohrer, J. Phys. Chem. B 102 (1998) 3216. [732] T. Miki, H. Yanagi, Langmuir 14 (1998) 3405. [733] M. Svetina, L. Colombi Ciacchi, O. Sbaizero, S. Meriani, A. De Vita, Acta Mater. 49 (2001) 2169. [734] G. Chiarelo, R. Barberi, E. Colavita, Appl. Surf. Sci. 99 (1996) 15. [735] Z. Chang, S. Piligkos, P.J. Moller, Phys. Rev. B 64 (2001) 165410/1. [736] J. Evans, B.E. Hayden, G. Lu, J. Chem. Soc., Faraday Trans. 92 (1996) 4733. [737] H.R. Sadeghi, V.E. Henrich, Appl. Surf. Sci. 19 (1984) 330. [738] B.E. Hayden, A. King, M.A. Newton, Chem. Phys. Lett. 269 (1997) 485. [739] F. Cosandey, P. Lu, Structure and Properties of Interfaces in Ceramics Symposium, Mater. Res. Soc. 1995, pp. 301±306. [740] M. Bowker, P. Stone, R. Bennett, N. Perkins, Surf. Sci. 497 (2002) 155. [741] S. Fischer, K.D. Schierbaum, W. GoÈpel, Sens. Actuators B 31 (1996) 13. [742] S. Fischer, F. Schneider, K.K. Schierbaum, Vacuum 47 (1996) 1149. [743] W.-x. Xu, K.D. Schierbaum, W. GoÈpel, J. Solid State Chem. 119 (1995) 237. [744] K. Tamura, U. Bardi, Y. Nihei, Surf. Sci. 216 (1989) 209. [745] K. Tamura, M. Owari, Y. Nihei, Bull. Chem. Soc. Jpn. 61 (1988) 1539. [746] K. Tamura, M. Kudo, M. Owari, Y. Nihei, Chem. Lett. (1986) 1921. [747] M. Marien, T. Wagner, D.L. Carroll, J. Marien, D.A. Bonnell, M. RuÈhle, MRS Bull. XXII (1997) 42. [748] P.J. Mùller, S.A. Komolov, E.F. Lazneva, Springer Ser. Surf. Sci. 33 (1993) 156. [749] M.C. Wu, P.J. Mùller, Chem. Phys. Lett. 171 (1990) 136. [750] M.C. Wu, P.J. Mùller, Surf. Sci. 235 (1990) 228. [751] D. Abriou, D. Gagnot, J. Jupille, F. Creuzet, Surf. Rev. Lett. 5 (1998) 387. [752] X. Lai, D.W. Goodman, J. Molec. Catal. 162 (2000) 33. [753] C.C. Chusuei, X. Lai, K.A. Davis, E.K. Bowers, J.P. Fackler, D.W. Goodman, Langmuir 17 (2001) 4113. [754] C. Xu, D.W. Goodman, Chem. Phys. Lett. 263 (1996) 13. [755] Z. Yang, R. Wu, D.W. Goodman, Phys. Rev. B 61 (2000) 14066. [756] S.C. Parker, A.W. Grant, V.A. Bondzie, C.T. Campbell, Surf. Sci. 441 (1999) 10. [757] V.A. Bondzie, S.C. Parker, C.T. Campbell, J. Vac. Sci. Technol. A 17 (1999) 1717. [758] N. Nilius, N. Ernst, H.J. Freund, Surf. Sci. 478 (2001) L327. [759] V.E. Henrich, R.L. Kurtz, Phys. Rev. B 23 (1981) 6280. [760] L.Q. Wang, K.F. Ferris, M.H. Engelhard, Surf. Sci. 440 (1999) 60. [761] V.V. Lobanov, Zh. Fiz. Khim. 70 (1996) 1440. [762] V.E. Henrich, G. Dresselhaus, H.J. Zeiger, Solid State Commun. 24 (1977) 623. [763] S. Eriksen, P.D. Naylor, R.G. Egdell, Spectrochim. Acta, Part 43A 12 (1987) 1535. [764] D.R. Baer, L.Q. Wang, A.N. Shultz, J.L. Daschbach, W.M. Hetherington III, M.H. Engelhard, in: Proceedings of the New Tech. Charact. Corros. Stress Corros., 1996, p. 73. [765] S. Krischok, O. Hofft, J. Gunster, J. Stultz, D.W. Goodman, V. Kempter, Surf. Sci. 495 (2001) 8. [766] J. Goniakowski, S. Bouette-Russo, C. Noguera, Surf. Sci. 284 (1993) 315. [767] J. Goniakowski, C. Noguera, Surf. Sci. 330 (1995) 337. [768] J. Goniakowski, M.J. Gillan, Surf. Sci. 350 (1996) 145. [769] V.V. Lobanov, Zh. Fiz. Khim. 70 (1996) 2267. [770] D. Vogtenhuber, R. Podloucky, J. Redinger, Surf. Sci. 404 (1998) 798. [771] F.J. Bustillo, E. RomaÂn, J.L. de Segovia, Vacuum 39 (1989) 659. [772] F.J. Bustillo, E. RomaÂn, J.L. de Segovia, Vacuum 41 (1990) 19. [773] P.B. Smith, S.L. Bernasek, Surf. Sci. 188 (1987) 241. [774] T.K. Sham, M.S. Lazarus, Chem. Phys. Lett. 68 (1979) 426. U. Diebold / Surface Science Reports 48 (2003) 53±229 229

[775] E. Kobayashi, K. Matsuda, G. Mizutani, S. Ushioda, Surf. Sci. 428 (1999) 294. [776] P. Fenter, L. Cheng, S. Rihs, M. Machesky, M.J. Bedzyk, N.C. Sturchio, J. Colloid Interf. Sci. 225 (2000) 154. [777] R. Hengerer, L. Kavan, B. Bolliger, M. Erbudak, M. GraÈtzel, Mater. Res. Soc. Symp. Proc. 623 (2000) 43. [778] L.M. Wu, Y.F. Zhang, Y. Li, J.Q. Li, L.X. Zhou, Jiegou Huaxue 18 (1999) 304. [779] J.A. Rodriguez, J. Hrbek, J. Dvorak, T. Jirsak, A. Maiti, Chem. Phys. Lett. 336 (2001) 377. [780] Y. Yamaguchi, H. Onishi, Y. Iwasawa, J. Chem. Soc., FaradayTrans. 91 (1995). [781] I.D. Cocks, Q. Guo, E.M. Williams, Surf. Sci. 390 (1997) 119. [782] Q. Guo, I. Cocks, E.M. Williams, J. Chem. Phys. 106 (1997) 2924. [783] P. Persson, A. Stashans, R. Bergstrom, S. Lunell, Int. J. Quantum Chem. 70 (1998) 1055.